devrag — Claude Code MCP server for Markdown search
This local RAG server indexes Markdown files and retrieves relevant passages from natural-language queries. Use it when an agent needs to search project docs without reading whole files.
Open on GitHubWhat devrag does
devrag creates a vector index from Markdown documents and exposes semantic search through MCP. An agent can narrow results by directory or filename pattern, then index, re-index, or remove documents. A separate CLI provides search and index-management commands.
Set file locations with document_patterns and choose the database location with db_path. Documents are indexed when the server starts, and list_documents returns the indexed files. The app is distributed as a single binary; it downloads its models on the first run, so that step needs network access. The macOS package includes libonnxruntime.dylib for CoreML acceleration, which must stay beside the devrag binary.
What it helps with
- find an API description in Markdown documentation
- limit document search by folder or filename pattern
- add a Markdown file to the index
- refresh the index for an edited document
- remove an outdated file from the index
Example prompts for the agent
- Find the JWT authentication approach in the project docs.
- Search for user endpoint details in
docs/apionly. - Add
./docs/api-spec.mdto the index.
How to install devrag
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about devrag
How do I install devrag in Claude Code?
Download and unpack the binary for your system. On macOS or Linux, make it executable and move it to /usr/local/bin/; then add a stdio server entry pointing to the binary in ~/.claude.json or .mcp.json.
What does devrag do?
It indexes Markdown files and retrieves relevant passages through semantic search. Its MCP tools also let you manage indexed documents.
Which files does devrag index?
It indexes Markdown files matched by document_patterns. The setting accepts directories and glob patterns, including recursive patterns.
