Open-source proxy that lets the Claude Code CLI run on Databricks Model Serving

2 points by vishalveera ↗ HN
It emulates the Claude Code backend locally and adds: – repo indexing + CLAUDE.md summaries – symbol search + cross-file references – git automation (diffs, commit, push policies) – MCP server orchestration (JSON-RPC tools) – prompt caching layer – optional Docker sandboxing – workspace tools for tests, tasks, diffs, file ops

Useful if you're doing LLM/AI engineering in environments where you need: – Databricks as your LLM runtime – private repo access – custom agent tools – auditability and local control

Repo: https://github.com/vishalveerareddy123/Lynkr Docs: https://vishalveerareddy123.github.io/Lynkr

Would love feedback, issues, or feature ideas from folks experimenting with LLM developer workflows.

1 comment

[ 18.6 ms ] story [ 850 ms ] thread
Claude Code is surprisingly strict about response shape, tool timing, and streaming semantics. Even small differences in Azure/Databricks responses cause silent failures.

Lynkr sits in the middle and enforces “Anthropic-ness” as a contract:

request rewriting

tool call lifecycle enforcement

streaming normalization

capability emulation when the backend can’t support something natively

Happy to explain any of the protocol quirks if people are curious.