
Claude Context MCP Server: Setup for Claude Code, Cursor and Other Apps
Lets Claude Code, Cursor and other coding assistants search your whole codebase by meaning, not just by file name.
Claude Context indexes a project folder and lets your assistant look up the code that matches a plain-English question. It runs on your computer as a local program. Zilliz, the company behind the Milvus vector database, makes it. It needs an embedding key (OpenAI by default) and a Zilliz Cloud key, so set up both accounts before you install it.
Good for
- Developers working in large repos where the assistant cannot read every file
- Claude Code and Cursor users who want fewer tokens spent on exploring
- Teams already running Milvus or Zilliz Cloud
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install claude-context MCP
Claude Code
- 1
Copy this prompt.
Add the claude-context MCP server to Claude Code. Run this command: claude mcp add claude-context -e OPENAI_API_KEY=<OPENAI_API_KEY> -e MILVUS_TOKEN=<MILVUS_TOKEN> -- npx -y @zilliz/claude-context-mcp Ask me for OPENAI_API_KEY, MILVUS_TOKEN and only put the values in that config, nowhere else. Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Node.js.
Built from the package data we hold. Not yet tested by us.
Tools
Builds a searchable index of a code folder so you can search it by meaning
Scans a project folder, splits the code into chunks and stores them in a vector database so they can be searched later. It runs in the background and can be forced to start over on a folder that is already indexed.
Try: “Index the project at /Users/me/code/shop so you can search it.”
Write ·
index_codebaseFinds code in an indexed project by describing what it does
Takes a plain-English question and returns the code chunks that match best, mixing keyword and meaning-based search. You can cap the number of results and filter by file type.
Try: “Find where we validate login tokens in /Users/me/code/shop.”
Network ·
search_codeShows whether a project is still being indexed and how far along it is
Reports progress while a folder is being indexed and whether indexing finished. Progress moves in coarse steps, so it can sit still during long embedding runs.
Try: “Is /Users/me/code/shop finished indexing?”
Read ·
get_indexing_statusRemoves the search index for a project
Deletes the stored search index for one folder. Your source files are not touched, and you can index the folder again afterwards.
Try: “Clear the search index for /Users/me/code/shop.”
Delete ·
clear_index
Read from the source code
Permissions
medium risk · 3 things to watch
Medium risk
It reads your project files, sends code chunks to your embedding provider and stores the index in a Zilliz Cloud or Milvus database. It can also delete the stored index for a folder.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
Code chunks leave your machine
Indexing sends your code to the embedding provider you pick and stores it in the vector database. Use Ollama and a local Milvus if the code must stay on your network. github.com
It can delete a folder's search index
The clear tool removes the stored index only, not your files. The index is a rebuildable copy of your code, so this is scored medium rather than high, but re-indexing a large repo costs embedding fees. Confirm the folder before it runs. github.com
Indexing a large repo costs money
Each indexed chunk is embedded by a paid API unless you use Ollama. Add ignore patterns for build output and vendored code. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only setting is documented. Search and status tools only read the index; indexing and clearing change it. |
|---|---|
| Talks to | Your embedding provider (OpenAI, VoyageAI, Gemini or OpenRouter; Ollama runs locally). Zilliz Cloud or your Milvus server |
| Sign-in and keys | Keys go in the server config as environment variables: an embedding provider key and a Zilliz Cloud API key. |
| Data retention | Code chunks and their vectors stay in your Zilliz Cloud or Milvus database until you clear the index. |
Questions
Do I need a Zilliz Cloud account?
Yes, or your own Milvus server. The README says the server needs a vector database, and its examples use a Zilliz Cloud API key. You can leave out the endpoint address because the server finds it from the token.
Which embedding providers does it support?
OpenAI (the default), VoyageAI, Gemini, OpenRouter and Ollama. Pick one with the provider setting and give the matching key. Ollama runs on your machine and needs no key.
Can I run it fully on my own machine?
The server itself always runs locally. For code to stay local too, you would use Ollama for embeddings and a Milvus server you host. The README documents Ollama and says Milvus can be local or cloud.
Which apps does it work with?
The README gives setup for Claude Code, Codex, Gemini CLI, Cursor, Windsurf, VS Code, Cline, Zed and more. Use the install box above for your app.
Can it change or delete my code?
No. It only reads your files to build the index. It can delete the stored search index for a folder, but never your source files.
Does it keep the index up to date?
Yes. It re-checks for changes in the background on a timer, and background sync can be turned off with a setting. An immediate re-sync by touching a trigger file only works if you set CLAUDE_CONTEXT_TRIGGER_WATCHER=true.
Is it official?
It is made by Zilliz, the company behind Milvus, and ships from its own repository. It is not made by Anthropic, OpenAI or Cursor.
Information
updated Nov 2, 2026
- Publisher
- Community
- Package
@zilliz/claude-context-mcp- Registry
- npm
- Transport
- stdio
- Auth
- API key
- License
- MIT
- Last commit
- Jul 14, 2026
- Open issues
- 81
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 2Hide
- github.com · github.com
- github.com · github.com