Second, a question. Computer Use and JSON mode are great for creating a quasi-API for legacy software which offers no integration possibilities. Can MCP better help with legacy software interactions, and if so, in what ways?
Probably, yes! You could imagine building an MCP server (integration) for a particular piece of legacy software, and inside that server, you could employ Computer Use to actually use and automate it.
The benefit would be that to the application connecting to your MCP server, it just looks like any other integration, and you can encapsulate a lot of the complexity of Computer Use under the hood.
If you explore this, we'd love to see what you come up with!
Will this be partially available from the Claude website for connections to other web services? E.g. could the GitHub server be called from https://claude.ai?
Any idea on timelines? I’d love to be able to have generation and tool use contained within a customer’s AWS account using bedrock. Ie I pass a single cdk that can interface with an exposed internet MCP service and an in-VPC service for sensitive data.
A few common use cases that I've been using is connecting a development database in a local docker container to Claude Desktop or any other MCP Client (e.g. an IDE assistant panel). I visualized the database layout in Claude Desktop and then create a Django ORM layer in my editor (which has MCP integration).
Zed editor had just announced support for MSP in some of their extensions, publishing an article showing some possible use cases/ideas: https://zed.dev/blog/mcp
Are there any resources for building the LLM side of MCP so we can use the servers with our own integration? Is there a specific schema for exposing MCP information to tool or computer use?
If you have specific questions, please feel free to start a discussion on the respective https://github.com/modelcontextprotocol discussion, and we are happy to help you with integrating MCP.
Ahh thanks! I was gonna say it's broken, but I now see that you're supposed to notice the sidebar changed and select one of the child pages. Would def recommend changing the sidebar link to that path instead of the index -- I would do it myself but couldn't find the sidebar in your doc repos within 5 minutes of looking.
Thanks for your hard work! "LSP for LLMs" is a fucking awesome idea
It's not exactly immutable, but any backwards incompatible changes would require a version bump.
We don't have a roadmap in one particular place, but we'll be populating GitHub Issues, etc. with all the stuff we want to get to! We want to develop this in the open, with the community.
Is it at least somewhat in sync with plans from Microsoft , OpenAI and Meta? And is it compatible with the current tool use API and computer use API that you’ve released?
From what I’ve seen, OpenAI attempted to solve the problem by partnering with an existing company that API-fys everything. This feels looks a more viable approach, if compared to effectively starting from scratch.
Is there any way to give a MCP server access for good? Trying out the demo it asked me every single time for permission which will be annoying for longer usage.
We do want to improve this over time, just trying to find the right balance between usability and security. Although MCP is powerful and we hope it'll really unlock a lot of potential, there are still risks like prompt injection and misconfigured/malicious servers that could cause a lot of damage if left unchecked.
I just want to say kudos for the design of the protocol. Seems inspired by https://langserver.org/ in all the right ways. Reading through it is a delight, there's so many tasteful little decisions.
One bit of constructive feedback: the TypeScript API isn't using the TypeScript type system to its fullest. For example, for tool providers, you could infer the type of a tool request handler's params from the json schema of the corresponding tool's input schema.
I guess that would be assuming that the model is doing constrained sampling correctly, such that it would never generate JSON that does not match the schema, which you might not want to bake into the reference server impl. It'd mean changes to the API too, since you'd need to connect the tool declaration and the request handler for that tool in order to connect their types.
This is a great idea! There's also the matter of requests' result types not being automatically inferred in the SDK right now, which would be great to fix.
Could I convince you to submit a PR? We'd love to include community contributions!
If you were willing to bring additional zod tooling or move to something like TypeBox (https://github.com/sinclairzx81/typebox), the json schema would be a direct derivation of the tools' input schemas in code.
The json-schema-to-ts npm package has a FromSchema type operator that converts the type of a json schema directly to the type of the values it describes. Zod and TypeBox are good options for users, but for the reference implementation I think a pure type solution would be better.
Superb work and super promising! I had wished for a protocol like this.
Is there a recommended resource for building MCP client? From what I've seen it just mentions Claude desktop & co are clients. SDK readme seems to cover it a bit but some examples could be great.
If you run into issues, feel free to open a discussion in the respective SDK repository and we are happy to help.
(I've been fairly successful in taking the spec documentation in markdown, an SDK and giving both to Claude and asking questions, but of course that requires a Claude account, which I don't want to assume)
I'm looking at integrating MCP with desktop app. The spec (https://spec.modelcontextprotocol.io/specification/basic/tra...) mentions "Clients SHOULD support stdio whenever possible.". The server examples seem to be mostly stdio as well. In the context of a sandboxed desktop app, it's often not practical to launch a server as subprocess because:
- sandbox restrictions of executing binaries
- needing to bundle binary leads to a larger installation size
Would it be reasonable to relax this restriction and provide both SSE/stdio for the default server examples?
Having broader support for SSE in the servers repository would be great. Maybe I can encourage you to open a PR or at least an issue.
I can totally see your concern about sandboxed app, particularly for flatpack or similar distribution methods. I see you already opened a discussion https://github.com/modelcontextprotocol/specification/discus..., so let's follow up there. I really appreciate the input.
A possible cheap win for servers would be to support the systemd "here's an fd number you get exec'ed with" model - that way server code that's only written to do read/write on a normal fd should be trivial to wire up to unix sockets, TCP sockets, etc.
(and then having a smol node/bun/go/whatever app that can sit in front of any server that handles stdio - or a listening socket for a server that can handle multiple clients - and translates the protocol over to SSE or websockets or [pick thing you want here] lets you support all such servers with a single binary to install)
Not that there aren't advantages to having such things baked into the server proper, but making 'writing a new connector that works at all' as simple as possible while still having access to multiple approaches to talk to it seems like something worthy of consideration.
[possibly I should've put this into the discussion, but I have to head out in a minute or two; anybody who's reading this and engaging over there should feel free to copy+paste anything I've said they think is relevant]
It seems extremely verbose. Why does the transport mechanism matter? Would have loved a protocol/standard about how best to organize/populate the context. I think MCP touches on that but has too much of other stuff for me.
In case of Claude Desktop App, I assume the decision which MCP-server's tool to use based on the end-user's query is done by Claude LLM using something like ReAct loop. Are the prompts and LLM-generated tokens involved inside "Protocol Handshake"-phase available for review?
1. The sampling documentation is confusing. "Sampling" means something very specific in statistics, and I'm struggling to see any connection between the term's typical usage and the usage here. Perhaps "prompt delegation" would be a more obvious term to use.
Another thing that's confusing about the sampling concept is that it's initiated by a server instead of a client, a reversal of how client/server interactions normally work. Without concrete examples, it's not obvious why or how a server might trigger such an exchange.
2. Some information on how resources are used would be helpful. How do resources get pulled into the context for queries? How are clients supposed to determine which resources are relevant? If the intention is that clients are to use resource descriptions to determine which to integrate into prompts, then that purpose should be more explicit.
Perhaps a bigger problem is that I don't see how clients are to take a resource's content into account when analyzing its relevance. Is this framework intentionally moving away from the practice of comparing content and query embeddings? Or is this expected to be done by indices maintained on the client?
I read through several of the top level pages, then SQLite, but still had no idea what was meant by "context" as it's a highly ambiguous word and is never mentioned with any concrete definition, example, or scope of capability that it is meant to imply.
After reading the Python server tutorial, it looks like there is some tool calling going on, in the old terminology. That makes more sense. But none of the examples seem to indicate what the protocol is, whether it's a RAG sort of thing, do I need to prompt, etc.
It would be nice to provide a bit more concrete info about capabilities and what the purposes is before getting into call diagrams. What do the arrows represent? That's more important to know than the order that a host talks to a server talks to a remote resource.
I think this is something that I really want and want to build a server for, but it's unclear to me how much more time I will have to invest before getting the basic information about it!
The gist of it is: you have an llm application such as Claude desktop. You want to have it interact (read or write) with some system you have. MCP solves this.
For example you can give the application the database schema as a “resource”, effectively saying; here is a bunch of text, do whatever you want with it during my chat with the llm. Or you can give the application a tool such as query my database. Now the model itself can decide when it wants to query (usually because you said: hey tell me what’s in the accounts table or something similar).
It’s “bring the things you care about” to any llm application with an mcp client
We definitely hope this will solve the NxM problem.
On tools specifically, we went back and forth about whether the other primitives of MCP ultimately just reduce to tool use, but ultimately concluded that separate concepts of "prompts" and "resources" are extremely useful to express different _intentions_ for server functionality. They all have a part to play!
I think this where the real question is for me. When I read about MCP, the topmost question in my mind is "Why isn't this just tool calling?" I had difficulty finding an answer to this. Below, you have someone else asking "Why not just use GraphQL?" And so on.
It would probably be helpful for many of your readers if you had a focused document that addressed specifically that motivating question, together with illustrated examples. What does MCP provide, and what does it intend to solve, that a tool calling interface or RPC protocol can't?
Yeah even I don't understand how it exactly solves the NXM problem (which translates to having M different prompts for N different llms. corerct me if I'm wrong please)
N (LLM clients/vendors) x M (tools/tool suppliers).
The N×M problem may simply be moved rather than solved:
- Instead of N×M direct integrations
- We now have N MCP client implementations
- M MCP server implementations
This feels similar to SOAP but might be more of a lower level protocol similar to HTTP itself. Hard to tell with the implementation examples being pretty subjective programs in python.
It seems to support your ask, as much as a protocol can. Having read all the docs and looked through some code, my mental model is:
- A host never talks to a server directly, only via a Client (which is presumably a human). The host has or is the LLM (app).
- A server only supplies context data (readonly), in the form of tool call, direct resource URL, or pre populated prompt. It can call back to a client directly, for example to request something from the hosts LLM.
- A client sits in the middle, representing the human in the loop. It manages the requests bidirectionally
It seems mostly modeled around the security boundaries, rather than just AI capabilities domains. The client is always in the loop, the host and server do not directly communicate.
How can an add on that works with arbitrary "servers" tell the difference between these two tools? Without being able to tell the difference you can't really build a generic way to ask for confirmation in the application that is using the server...
{
name: "create_directory",
description:
"Create a new directory or ensure a directory exists. Can create multiple " +
"nested directories in one operation. If the directory already exists, " +
"this operation will succeed silently. Perfect for setting up directory " +
"structures for projects or ensuring required paths exist. Only works within allowed directories.",
inputSchema: zodToJsonSchema(CreateDirectoryArgsSchema) as ToolInput,
},
{
name: "list_directory",
description:
"Get a detailed listing of all files and directories in a specified path. " +
"Results clearly distinguish between files and directories with [FILE] and [DIR] " +
"prefixes. This tool is essential for understanding directory structure and " +
"finding specific files within a directory. Only works within allowed directories.",
inputSchema: zodToJsonSchema(ListDirectoryArgsSchema) as ToolInput,
},
Great work on the protocol!!
I am looking for some examples of creating my own custom client with the Anthropic API leveraging MCP, but I could not find any. Pretty much want to understand how Claude Desktop is integrating with MCP Server along with Anthropic API
Can you provide some pointers about the integration?
e.g.
At first glance it seems to be a proposed standard interface and protocol for describing and offering an external system to the function calling faculity of an LLM.
> had no idea what was meant by "context" as it's a highly ambiguous word and is never mentioned with any concrete definition
(forgive me if you know this and are asking a different question, but:)
I don't know how familiar you are with LLMs, but "context" used in that context generally has the pretty clear meaning of "the blob of text you give in between (the text of) the system prompt and (the text of) the user prompt"[1], which acts as context for the user's request (hence the name). Very often this is the conversation history in chatbot-style LLMs, but it can include stuff like the content of text files you're working with, or search/function results.
[1] If you want to be pedantic, technically each instance of "text" should say "tokens" there, and the maximum "context" length includes the length of both prompts.
The result that MCP server returned will be transfer to MCP host(Claude, IDEs, Tools), there are some privacy issues because the process is automatic after one-time permission provided.
For instance, when there is something wrong for MCP host, it query all data from database and transfer it to host, all data will be leaked.
It's hard to totally prevent this kind of problem when interacting with local data, But, Is there some actions to prevent this kind of situations for MCP?
Your concerns are very valid. This is partly why right now, in Claude Desktop, it's not possible to grant permission permanently. The most you can do is "Allow for this chat," which applies to one tool from one server at a time.
For Rust, could one leverage the type + docs system to create such a server? I didn't delve into the details but one of the issues of Claude is that it has no knowledge of the methods that are available to it (vs LSP). Will creating such a server make it able to do informed suggestions?
I'd love to develop some MCP servers, but I just learned that Claude Desktop doesn't support Linux. Are there any good general-purpose MCP clients that I can test against? Do I have to write my own?
(Closest I can find is zed/cody but those aren't really general purpose)
You guys need a professional documentation person on your team, one that specializes in only writing documentation. I say this because the existing documentation is a confusing mess. This is going to cause all kinds of problems purely because it is weakly explained, and I see incorrect usage of words all over. Even the very beginning definitions of client, host and server are nonstandard.
Any ideas on how the concepts here will mesh with the recently released Microsoft.Extensions.AI library released by MS for .NET, that is also supposed to make it easy to work with different models in a standardized way?
this is really cool stuff. I just started to write a server and I have a few questions. Not sure if HN is the right place, so where would you suggest to ask them?
Anyway, if there is no place yet, my questions are:
- In the example https://modelcontextprotocol.io/docs/first-server/python , what is the difference between read_resources and call_tool. In both cases the call the fetch_weather function. Would be nice to have that explained better.
I implemented in my own server only the call_tool function and Claude seems to be able to call it.
- Where is inputSchema of Tool specified in the docs? It would be nice if inputSchema would be explained a bit better. For instance how can I make a list of strings field that has a default value.
- How can i view the output of logger? It would be nice to see somewhere an example on how to check the logs. I log some stuff with logger.info and logger.error but I have no clue where I can actually look at it. My work around now is to log to a local file and tail if..
General feedback
- PLEASE add either automatic reload of server (hard) or a reload button in the app (probably easier). Its really disrupting to the flow when you have ot restart the app on any change.
- Claude Haiku never calls the tools. It just tells me it can't do it. Sonnet can do it but is really slow.
- The docs are really really version 0.1 obviously :-) Please put some focus on it...
I have a case in mind where I would like to connect to multiple databases. Does the integration endpoint specification in claude_desktop_config.json allow us to pass some description so as to differentiate different databases? How?
I can see where you're going with this and I can understand why you don't want to get into authorization, but if you're going to be encouraging tool developers to spin up json-rpc servers I hope you have some kind of plan for authorization otherwise you're encouraging a great way to break security models. Just because it's local doesn't mean it's secure. This protocol is dead the moment it becomes an attack vector.
Did I misunderstand, or does it not seem to have support for user authentication? It seems your operating model is that the MCP server is, during installation time, configured authentication for the underlying service. This is fine for non-serious use cases such as weather forecast querying, or for small-scale situations where only a couple of people have access to an LLM that's connected to the MCP server. But in an enterprise setting, there are thousands of people, whose level of access to the service behind the MCP server, differs. I think the MCP server needs a way to know the identity of the human behind the LLM, so that it can perform appropriate authentication and authorization.
Not sure I understand your point. If it's your client / server, you are controlling how they interact, by implementing the necessaries according to the protocol.
If you're writing an LSP for a language, you're implementing the necessaries according to the protocol (when to show errors, inlay hints, code fixes, etc.) - it's not deciding on its own.
Even if I could make use of it, I wouldn't, because I don't write proprietary code that only works on one AI Service Provider. I use only LangChain so that all of my code can be used with any LLM.
My app has a simple drop down box where users can pick whatever LLM they want to to use (OpenAI, Perplexity, Gemini, Anthropic, Grok, etc)
However if they've done something worthy of putting into LangChain, then I do hope LangChain steals the idea and incorporates it so that all LLM apps can use it.
It's an open protocol; where did you get the idea that it would only work with Claude? You can implement it for whatever you want - I'm sure langchain folks are already working on something to accommodate it
Once fully adopted by at least 3 other companies I'll consider it a standard, and would consider it yes, if it solved a problem I have, which it does not.
Lots of companies open source some of their internal code, then say it's "officially a protocol now" that anyone can use, and then no one else ever uses it.
If they have new "tools" that's great however, but only as long as they can be used in LangChain independent of any "new protocol".
I’m glad they're pushing for standards here, literally everyone has been writing their own integrations and the level of fragmentation (as they also mention) and repetition going into building the infra around agents is super high.
We’re building an in terminal coding agent and our next step was to connect to external services like sentry and github where we would also be making a bespoke integration or using a closed source provider. We appreciate that they have mcp integrations already for those services. Thanks Anthropic!
I've been implementing a lot of this exact stuff over the past month, and couldn't agree more. And they even typed the python SDK -- with pydantic!! An exciting day to be an LLM dev, that's for sure. Will be immediately switching all my stuff to this (assuming it's easy to use without their starlette `server` component...)
As someone building a client which needs to sync with a local filesystem (repo) and database, I cannot emphasize how wonderful it is that there is a push to standardize. We're going to implement this for https://srcbook.com
Hmm I like the idea of providing a unified interface to all LLMs to interact with outside data.
But I don't really understand why this is local only. It would be a lot more interesting if I could connect this to my github in the web app and claude automatically has access to my code repositories.
I guess I can do this for my local file system now?
I also wonder if I build an LLM powered app, and currently simply to RAG and then inject the retrieved data into my prompts, should this replace it? Can I integrate this in a useful way even?
The use case of on your machine with your specific data, seems very narrow to me right now, considering how many different context sources and use cases there are.
> It would be a lot more interesting if I could connect this to my github in the web app and claude automatically has access to my code repositories.
From the link:
> To help developers start exploring, we’re sharing pre-built MCP servers for popular enterprise systems like Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer.
However, it's not quite a complete story yet. Remote connections introduce a lot more questions and complexity—related to deployment, auth, security, etc. We'll be working through these in the coming weeks, and would love any and all input!
Will you also create some info on how other LLM providers can integrate this? So far it looks like it's mostly a protocol to integrate with anthropic models/desktop client. That's not what I thought of when I read open-source.
It would be a lot more interesting to write a server for this if this allowed any model to interact with my data. Everyone would benefit from having more integration and you (anthropic) still would have the advantage of basically controlling the protocol.
Note that both Sourcegraph's Cody and the Zed editor support MCP now. They offer other models besides Claude in their respective application.
The Model Context Protocol initial release aims to solve the N-to-M relation of LLM applications (mcp clients) and context providers (mcp servers). The application is free to choose any model they want. We carefully designed the protocol such that it is model independent.
LLM applications just means chat applications here though right?
This doesn't seem to cover use cases of more integrated software. Like a typical documentation RAG chatbot.
Local only solves a lot of problems. Our infrastructure does tend to assume that data and credentials are on a local computer - OAuth is horribly complex to set up and there's no real benefit to messing with that when local works fine.
I'm honestly happy with them starting local-first, because... imagine what it would look like if they did the opposite.
> It would be a lot more interesting if I could connect this to my github in the web app and claude automatically has access to my code repositories.
In which case the "API" would be governed by a contract between Anthropic and Github, to which you're a third party (read: sharecropper).
Interoperability on the web has already been mostly killed by the practice of companies integrating with other companies via back-channel deals. You are either a commercial partner, or you're out of the playground and no toys for you. Them starting locally means they're at least reversing this trend a bit by setting a different default: LLMs are fine to integrate with arbitrary code the user runs on their machine. No need to sign an extra contact with anyone!
I can see the value of something like DSPy where there is some higher level abstractions in wiring together a system of llms.
But this seems like an abstraction that doesn't really offer much besides "function calling but you use our python code".
I see the value of language server protocol but I don't see the mapping to this piece of code.
That's actually negative value if you are integrating into an existing software system or just you know... exposing functions that you've defined vs remapping functions you've defined into this intermediate abstraction.
The secret sauce part is the useful part -- the local vector store. Anthropic is probably not going to release that without competitive pressure. Meanwhile this helps Anthropic build an ecosystem.
When you think about it, function calling needs its own local state (embedded db) to scale efficiently on larger contexts.
I'd like to see all this become open source / standardized.
im not sure what you mean - the embedding model is independent of the embeddings themselves. Once generated, the embeddings and vector store should exist 100% locally and thus not part of any secret sauce
If integrations are required to unlock value, then the platform with the most prebuilt integrations wins.
The bulk of mass adopters don't have the in-house expertise or interest in building their own. They want turnkey.
No company can build integrations, at scale, more quickly itself than an entire community.
If Anthropic creates an integration standard and gets adoption, then it either at best has a competitive advantage (first mover and ownership of the standard) or at worst prevents OpenAI et al. from doing the same to it.
(Also, the integration piece is the necessary but least interesting component of the entire system. Way better to commodify it via standard and remove it as a blocker to adoption)
My team and I have a desktop product with a very similar architecture (a central app+UI with a constellation of local servers providing functions and data to models for local+remote context)
If this protocol gets adoption we'll probably add compatibility.
Which would bring MCP to local models like LLama 3 as well as other cloud providers competitors like OpenAI, etc
After a quick look it seemed to me like they're trying to standardize on how clients call servers, which nobody needs, and nobody is going to use. However if they have new Tools that can be plugged into my LangChain stuff, that will be great, and I can use that, but I have no place for any new client/server models.
So they want an open protocol, and instead of say collaborating with other people that provide models like Google, Microsoft, Mistral, Cohere and the opensource community, they collaborate with an editor team. Quite the protocol. Why should Microsoft implement this? If they implement their own protocol, they win. Why should Google implement this? If they implement their own protocol, they win too. Both giants have way more apps and reach in inside businesses than Anthropic can wish.
One thing I dont understand.. does this rely on vector embeddings? Or how does the AI interact with the data? The example is a sqllite satabase with prices, and it shows claude being asked to give the average price and to suggest pricing optimizations.
So does the entire db get fed into the context? Or is there another layer in between. What if the database is huge, and you want to ask the AI for the most expensive or best selling items? With RAG that was only vaguely possible and didnt work very well.
It never accidentally deletes anything? Or I guess you give it read only access? It is querying it through this API and some adapter built for it, or the file gets sent through the API, they recognize it is sqllite and load it on their end?
This is about tool usage - the thing where an LLM can be told "if you want to run a SQL query, say <sql>select * from repos</sql> - the code harness well then spot that tag, run the query for you and return the results to you in a chat message so you can use them to help answer a question or continue generating text".
The Model Context server is similar to what we've built at Spice, but we've focused on databases and data systems. Overall, standards are good. Perhaps we can implement MCP as a data connector and tool.
I would love to integrate this into my platform of tools for AI models, Toolhouse [1], but I would love to understand the adoption of this protocol, especially as it seems to only work with one foundational model.
I love how they’re pretending to be champions of open source while leaving this gem in their terms of use
“””
You may not access or use, or help another person to access or use, our Services in the following ways:
…
To develop any products or services that compete with our Services, including to develop or train any artificial intelligence or machine learning algorithms or models.
“””
OpenAI says, "[You may not] Use Output to develop models that compete with OpenAI." That feels more narrow than Anthropic's blanket ban on any machine learning development.
Presumably this doesn't apply to the standard being released here, nor any of its implementations made available. Each of these appears to have its own permissible license.
I think open-sourcing your tech for the common person while leaving commercial use behind a paywall or even just against terms is completely acceptable, no?
Building something for this at surferprotocol [dot] org. Imo not every company will expose API's for easily exporting data from their platforms (linkedin, imessage, etc), so devs have to build these themselves
I see a good number of comments that seem skeptical or confused about what's going on here or what the value is.
One thing that some people may not realize is that right now there's a MASSIVE amount of effort duplication around developing something that could maybe end up looking like MCP. Everyone building an LLM agent (or pseudo-agent, or whatever) right now is writing a bunch of boilerplate for mapping between message formats, tool specification formats, prompt templating, etc.
Now, having said that, I do feel a little bit like there's a few mistakes being made by Anthropic here. The big one to me is that it seems like they've set the scope too big. For example, why are they shipping standalone clients and servers rather than client/server libraries for all the existing and wildly popular ways to fetch and serve HTTP? When I've seen similar mistakes made (e.g. by LangChain), I assume they're targeting brand new developers who don't realize that they just want to make some HTTP calls.
Another thing that I think adds to the confusion is that, while the boilerplate-ish stuff I mentioned above is annoying, what's REALLY annoying and actually hard is generating a series of contexts using variations of similar prompts in response to errors/anomalies/features detected in generated text. IMO this is how I define "prompt engineering" and it's the actual hard problem we have to solve. By naming the protocol the Model Context Protocol, I assumed they were solving prompt engineering problems (maybe by standardizing common prompting techniques like ReAct, CoT, etc).
The issue isn’t with who’s hosting, it’s that their SDKs don’t clearly integrate with existing HTTP servers regardless of who’s hosting them. I mean integrate at the source level, of course they could integrate via HTTP call.
Your point about boilerplate is key, and it’s why I think MCP could work well despite some of the concerns raised. Right now, so many of us are writing redundant integrations or reinventing the same abstractions for tool usage and context management. Even if the first iteration of MCP feels broad or clunky, standardizing this layer could massively reduce friction over time.
Regarding the standalone servers, I suspect they’re aiming for usability over elegance in the short term. It’s a classic trade-off: get the protocol in people’s hands to build momentum, then refine the developer experience later.
I don't see I or any other developer would abandon their homebrew agent implementation for a "standard" which isn't actually a standard yet.
I also don't see any of that implementation as "boilerplate". Yes there's a lot of similar code being written right now but that's healthy co-evolution. If you have a look at the codebases for Langchain and other LLM toolkits you will realize that it's a smarter bet to just roll your own for now.
You've definitely identified the main hurdle facing LLM integration right now and it most definitely isn't a lack of standards. The issue is that the quality of raw LLM responses falls apart in pretty embarrassing ways. It's understood by now that better prompts cannot solve these problems. You need other error-checking systems as part of your pipeline.
The AI companies are interested in solving these problems but they're unable to. Probably because their business model works best if their system is just marginally better than their competitor.
really great to see some standards emerging. i'd love to see something like mindsdb wired up to support this protocol and get a bunch of stuff out of the box.
Tangential question: Is there any LLM which is capable of preserving the context through many sessions, so it doesn't have to upload all my context every time?
In an ideal world gemini (or any other 1M token context model) would have an internal 'save snapshot' option so one could resume a blank conversation after 'priming' the internal state (activations) with the whole code base.
282 comments
[ 2.8 ms ] story [ 280 ms ] threadSecond, a question. Computer Use and JSON mode are great for creating a quasi-API for legacy software which offers no integration possibilities. Can MCP better help with legacy software interactions, and if so, in what ways?
The benefit would be that to the application connecting to your MCP server, it just looks like any other integration, and you can encapsulate a lot of the complexity of Computer Use under the hood.
If you explore this, we'd love to see what you come up with!
And the GitHub repo: https://github.com/modelcontextprotocol
Internally we have seen people experiment with a wide variety of different integrations from reading data files to managing their Github repositories through Claude using MCP. Alex's post https://x.com/alexalbert__/status/1861079762506252723 has some good examples. Alternatively please take a look at https://github.com/modelcontextprotocol/servers for a set of servers we found useful.
https://github.com/anaisbetts/mcp-youtube
Claude doesn't support YouTube summaries. I thought that was annoying! So I added it myself, instead of having to hope Anthropic would do it
In addition, I recommend looking at the specification documentation at https://spec.modelcontextprotocol.io. This should give you a good overview of how to implement a client. If you are looking to see an implemented open source client, Zed implements an MCP client: https://github.com/zed-industries/zed/tree/main/crates/conte...
If you have specific questions, please feel free to start a discussion on the respective https://github.com/modelcontextprotocol discussion, and we are happy to help you with integrating MCP.
Is it versioned? ie. does this release constitute an immutable protocol for the time being?
Thanks for your hard work! "LSP for LLMs" is a fucking awesome idea
It's not exactly immutable, but any backwards incompatible changes would require a version bump.
We don't have a roadmap in one particular place, but we'll be populating GitHub Issues, etc. with all the stuff we want to get to! We want to develop this in the open, with the community.
From what I’ve seen, OpenAI attempted to solve the problem by partnering with an existing company that API-fys everything. This feels looks a more viable approach, if compared to effectively starting from scratch.
One bit of constructive feedback: the TypeScript API isn't using the TypeScript type system to its fullest. For example, for tool providers, you could infer the type of a tool request handler's params from the json schema of the corresponding tool's input schema.
I guess that would be assuming that the model is doing constrained sampling correctly, such that it would never generate JSON that does not match the schema, which you might not want to bake into the reference server impl. It'd mean changes to the API too, since you'd need to connect the tool declaration and the request handler for that tool in order to connect their types.
Could I convince you to submit a PR? We'd love to include community contributions!
Is there a recommended resource for building MCP client? From what I've seen it just mentions Claude desktop & co are clients. SDK readme seems to cover it a bit but some examples could be great.
The best starting point are the respective client parts in the SDK: https://github.com/modelcontextprotocol/typescript-sdk/tree/... and https://github.com/modelcontextprotocol/python-sdk/tree/main..., as well as the official specification documentation at https://spec.modelcontextprotocol.io.
If you run into issues, feel free to open a discussion in the respective SDK repository and we are happy to help.
(I've been fairly successful in taking the spec documentation in markdown, an SDK and giving both to Claude and asking questions, but of course that requires a Claude account, which I don't want to assume)
I'm looking at integrating MCP with desktop app. The spec (https://spec.modelcontextprotocol.io/specification/basic/tra...) mentions "Clients SHOULD support stdio whenever possible.". The server examples seem to be mostly stdio as well. In the context of a sandboxed desktop app, it's often not practical to launch a server as subprocess because:
- sandbox restrictions of executing binaries
- needing to bundle binary leads to a larger installation size
Would it be reasonable to relax this restriction and provide both SSE/stdio for the default server examples?
I can totally see your concern about sandboxed app, particularly for flatpack or similar distribution methods. I see you already opened a discussion https://github.com/modelcontextprotocol/specification/discus..., so let's follow up there. I really appreciate the input.
(and then having a smol node/bun/go/whatever app that can sit in front of any server that handles stdio - or a listening socket for a server that can handle multiple clients - and translates the protocol over to SSE or websockets or [pick thing you want here] lets you support all such servers with a single binary to install)
Not that there aren't advantages to having such things baked into the server proper, but making 'writing a new connector that works at all' as simple as possible while still having access to multiple approaches to talk to it seems like something worthy of consideration.
[possibly I should've put this into the discussion, but I have to head out in a minute or two; anybody who's reading this and engaging over there should feel free to copy+paste anything I've said they think is relevant]
In case of Claude Desktop App, I assume the decision which MCP-server's tool to use based on the end-user's query is done by Claude LLM using something like ReAct loop. Are the prompts and LLM-generated tokens involved inside "Protocol Handshake"-phase available for review?
1. The sampling documentation is confusing. "Sampling" means something very specific in statistics, and I'm struggling to see any connection between the term's typical usage and the usage here. Perhaps "prompt delegation" would be a more obvious term to use.
Another thing that's confusing about the sampling concept is that it's initiated by a server instead of a client, a reversal of how client/server interactions normally work. Without concrete examples, it's not obvious why or how a server might trigger such an exchange.
2. Some information on how resources are used would be helpful. How do resources get pulled into the context for queries? How are clients supposed to determine which resources are relevant? If the intention is that clients are to use resource descriptions to determine which to integrate into prompts, then that purpose should be more explicit.
Perhaps a bigger problem is that I don't see how clients are to take a resource's content into account when analyzing its relevance. Is this framework intentionally moving away from the practice of comparing content and query embeddings? Or is this expected to be done by indices maintained on the client?
After reading the Python server tutorial, it looks like there is some tool calling going on, in the old terminology. That makes more sense. But none of the examples seem to indicate what the protocol is, whether it's a RAG sort of thing, do I need to prompt, etc.
It would be nice to provide a bit more concrete info about capabilities and what the purposes is before getting into call diagrams. What do the arrows represent? That's more important to know than the order that a host talks to a server talks to a remote resource.
I think this is something that I really want and want to build a server for, but it's unclear to me how much more time I will have to invest before getting the basic information about it!
The gist of it is: you have an llm application such as Claude desktop. You want to have it interact (read or write) with some system you have. MCP solves this.
For example you can give the application the database schema as a “resource”, effectively saying; here is a bunch of text, do whatever you want with it during my chat with the llm. Or you can give the application a tool such as query my database. Now the model itself can decide when it wants to query (usually because you said: hey tell me what’s in the accounts table or something similar).
It’s “bring the things you care about” to any llm application with an mcp client
It's not introducing new capabilities, just solving the NxM problem, hopefully leading to more tools being written.
(At least that's how I understand this. Am I far off?)
On tools specifically, we went back and forth about whether the other primitives of MCP ultimately just reduce to tool use, but ultimately concluded that separate concepts of "prompts" and "resources" are extremely useful to express different _intentions_ for server functionality. They all have a part to play!
It would probably be helpful for many of your readers if you had a focused document that addressed specifically that motivating question, together with illustrated examples. What does MCP provide, and what does it intend to solve, that a tool calling interface or RPC protocol can't?
The N×M problem may simply be moved rather than solved:
This feels similar to SOAP but might be more of a lower level protocol similar to HTTP itself. Hard to tell with the implementation examples being pretty subjective programs in python.https://github.com/modelcontextprotocol/servers/blob/main/sr...
How can an add on that works with arbitrary "servers" tell the difference between these two tools? Without being able to tell the difference you can't really build a generic way to ask for confirmation in the application that is using the server...
import anthropic
client = anthropic.Anthropic()
response = client.messages.create( model="claude-3-5-sonnet-20241022", max_tokens=1024, [mcp_server]=... ## etc.? ... )
(forgive me if you know this and are asking a different question, but:)
I don't know how familiar you are with LLMs, but "context" used in that context generally has the pretty clear meaning of "the blob of text you give in between (the text of) the system prompt and (the text of) the user prompt"[1], which acts as context for the user's request (hence the name). Very often this is the conversation history in chatbot-style LLMs, but it can include stuff like the content of text files you're working with, or search/function results.
[1] If you want to be pedantic, technically each instance of "text" should say "tokens" there, and the maximum "context" length includes the length of both prompts.
For instance, when there is something wrong for MCP host, it query all data from database and transfer it to host, all data will be leaked.
It's hard to totally prevent this kind of problem when interacting with local data, But, Is there some actions to prevent this kind of situations for MCP?
(Closest I can find is zed/cody but those aren't really general purpose)
this is really cool stuff. I just started to write a server and I have a few questions. Not sure if HN is the right place, so where would you suggest to ask them?
Anyway, if there is no place yet, my questions are:
- In the example https://modelcontextprotocol.io/docs/first-server/python , what is the difference between read_resources and call_tool. In both cases the call the fetch_weather function. Would be nice to have that explained better. I implemented in my own server only the call_tool function and Claude seems to be able to call it.
- Where is inputSchema of Tool specified in the docs? It would be nice if inputSchema would be explained a bit better. For instance how can I make a list of strings field that has a default value.
- How can i view the output of logger? It would be nice to see somewhere an example on how to check the logs. I log some stuff with logger.info and logger.error but I have no clue where I can actually look at it. My work around now is to log to a local file and tail if..
General feedback
- PLEASE add either automatic reload of server (hard) or a reload button in the app (probably easier). Its really disrupting to the flow when you have ot restart the app on any change.
- Claude Haiku never calls the tools. It just tells me it can't do it. Sonnet can do it but is really slow.
- The docs are really really version 0.1 obviously :-) Please put some focus on it...
Overall, awesome work!
Thanks
I'm looking at a PostgreSQL integration here: https://github.com/modelcontextprotocol/servers/tree/main/sr...
I have a case in mind where I would like to connect to multiple databases. Does the integration endpoint specification in claude_desktop_config.json allow us to pass some description so as to differentiate different databases? How?
MCP: I've gotten 2,415 times smarter since then.
https://www.darinkishore.com/posts/mcp/
https://unrollnow.com/status/1861079762506252723
For example:
When and how should notifications be sent and how should they be handled?
---
It's a lot more like LSP.
If you're writing an LSP for a language, you're implementing the necessaries according to the protocol (when to show errors, inlay hints, code fixes, etc.) - it's not deciding on its own.
My app has a simple drop down box where users can pick whatever LLM they want to to use (OpenAI, Perplexity, Gemini, Anthropic, Grok, etc)
However if they've done something worthy of putting into LangChain, then I do hope LangChain steals the idea and incorporates it so that all LLM apps can use it.
Lots of companies open source some of their internal code, then say it's "officially a protocol now" that anyone can use, and then no one else ever uses it.
If they have new "tools" that's great however, but only as long as they can be used in LangChain independent of any "new protocol".
We’re building an in terminal coding agent and our next step was to connect to external services like sentry and github where we would also be making a bespoke integration or using a closed source provider. We appreciate that they have mcp integrations already for those services. Thanks Anthropic!
https://sourcegraph.com/blog/cody-supports-anthropic-model-c...
I guess I can do this for my local file system now?
I also wonder if I build an LLM powered app, and currently simply to RAG and then inject the retrieved data into my prompts, should this replace it? Can I integrate this in a useful way even?
The use case of on your machine with your specific data, seems very narrow to me right now, considering how many different context sources and use cases there are.
From the link:
> To help developers start exploring, we’re sharing pre-built MCP servers for popular enterprise systems like Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer.
I guess the reason for this local focus is, that it's otherwise hard to provide access to local files. Which is a decently large use-case.
Still it feels a bit complicated to me.
However, it's not quite a complete story yet. Remote connections introduce a lot more questions and complexity—related to deployment, auth, security, etc. We'll be working through these in the coming weeks, and would love any and all input!
It would be a lot more interesting to write a server for this if this allowed any model to interact with my data. Everyone would benefit from having more integration and you (anthropic) still would have the advantage of basically controlling the protocol.
The Model Context Protocol initial release aims to solve the N-to-M relation of LLM applications (mcp clients) and context providers (mcp servers). The application is free to choose any model they want. We carefully designed the protocol such that it is model independent.
Here's one for performing GitHub actions: https://cookbook.openai.com/examples/chatgpt/gpt_actions_lib...
> It would be a lot more interesting if I could connect this to my github in the web app and claude automatically has access to my code repositories.
In which case the "API" would be governed by a contract between Anthropic and Github, to which you're a third party (read: sharecropper).
Interoperability on the web has already been mostly killed by the practice of companies integrating with other companies via back-channel deals. You are either a commercial partner, or you're out of the playground and no toys for you. Them starting locally means they're at least reversing this trend a bit by setting a different default: LLMs are fine to integrate with arbitrary code the user runs on their machine. No need to sign an extra contact with anyone!
I can see the value of something like DSPy where there is some higher level abstractions in wiring together a system of llms.
But this seems like an abstraction that doesn't really offer much besides "function calling but you use our python code".
I see the value of language server protocol but I don't see the mapping to this piece of code.
That's actually negative value if you are integrating into an existing software system or just you know... exposing functions that you've defined vs remapping functions you've defined into this intermediate abstraction.
When you think about it, function calling needs its own local state (embedded db) to scale efficiently on larger contexts.
I'd like to see all this become open source / standardized.
If integrations are required to unlock value, then the platform with the most prebuilt integrations wins.
The bulk of mass adopters don't have the in-house expertise or interest in building their own. They want turnkey.
No company can build integrations, at scale, more quickly itself than an entire community.
If Anthropic creates an integration standard and gets adoption, then it either at best has a competitive advantage (first mover and ownership of the standard) or at worst prevents OpenAI et al. from doing the same to it.
(Also, the integration piece is the necessary but least interesting component of the entire system. Way better to commodify it via standard and remove it as a blocker to adoption)
If this protocol gets adoption we'll probably add compatibility.
Which would bring MCP to local models like LLama 3 as well as other cloud providers competitors like OpenAI, etc
We've been keeping quiet, but I'd be happy to chat more if you want to email me (also in bio)
So does the entire db get fed into the context? Or is there another layer in between. What if the database is huge, and you want to ask the AI for the most expensive or best selling items? With RAG that was only vaguely possible and didnt work very well.
Sorry I am a bit new but trying to learn more.
This is about tool usage - the thing where an LLM can be told "if you want to run a SQL query, say <sql>select * from repos</sql> - the code harness well then spot that tag, run the query for you and return the results to you in a chat message so you can use them to help answer a question or continue generating text".
[1] https://github.com/spiceai/spiceai
[1] https://toolhouse.AI
Would love to chat with you if you are open about possible collab.
I am frank [at] glama.ai
“”” You may not access or use, or help another person to access or use, our Services in the following ways: … To develop any products or services that compete with our Services, including to develop or train any artificial intelligence or machine learning algorithms or models. “””
Now let's see a similar abstraction on the client side - a unified way of connecting your assistant to Slack, Discord, Telegram, etc.
One thing that some people may not realize is that right now there's a MASSIVE amount of effort duplication around developing something that could maybe end up looking like MCP. Everyone building an LLM agent (or pseudo-agent, or whatever) right now is writing a bunch of boilerplate for mapping between message formats, tool specification formats, prompt templating, etc.
Now, having said that, I do feel a little bit like there's a few mistakes being made by Anthropic here. The big one to me is that it seems like they've set the scope too big. For example, why are they shipping standalone clients and servers rather than client/server libraries for all the existing and wildly popular ways to fetch and serve HTTP? When I've seen similar mistakes made (e.g. by LangChain), I assume they're targeting brand new developers who don't realize that they just want to make some HTTP calls.
Another thing that I think adds to the confusion is that, while the boilerplate-ish stuff I mentioned above is annoying, what's REALLY annoying and actually hard is generating a series of contexts using variations of similar prompts in response to errors/anomalies/features detected in generated text. IMO this is how I define "prompt engineering" and it's the actual hard problem we have to solve. By naming the protocol the Model Context Protocol, I assumed they were solving prompt engineering problems (maybe by standardizing common prompting techniques like ReAct, CoT, etc).
Regarding the standalone servers, I suspect they’re aiming for usability over elegance in the short term. It’s a classic trade-off: get the protocol in people’s hands to build momentum, then refine the developer experience later.
I also don't see any of that implementation as "boilerplate". Yes there's a lot of similar code being written right now but that's healthy co-evolution. If you have a look at the codebases for Langchain and other LLM toolkits you will realize that it's a smarter bet to just roll your own for now.
You've definitely identified the main hurdle facing LLM integration right now and it most definitely isn't a lack of standards. The issue is that the quality of raw LLM responses falls apart in pretty embarrassing ways. It's understood by now that better prompts cannot solve these problems. You need other error-checking systems as part of your pipeline.
The AI companies are interested in solving these problems but they're unable to. Probably because their business model works best if their system is just marginally better than their competitor.
In an ideal world gemini (or any other 1M token context model) would have an internal 'save snapshot' option so one could resume a blank conversation after 'priming' the internal state (activations) with the whole code base.