Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@ragd-mcp find where the API rate limiting is implemented"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP (Model Context Protocol) adapter that exposes ragd's
POST /query endpoint as a query_codebase tool, so Claude Code can call
it for fuzzy/semantic codebase search alongside its built-in Grep/Glob.
ragd speaks plain REST/JSON, not MCP, so this exists purely to bridge the two protocols. It's a single tool, resolved automatically:
If a resource name isn't given, it looks up ragd's /resources list
and picks whichever registered resource's directory contains the
current working directory (deepest match wins).
Otherwise it queries the given resource directly.
ragd running and reachable (default http://localhost:20250,
override with RAGD_URL)
Related MCP server: RAGex
Run from inside this repo:
claude mcp add --transport stdio ragd --scope user -- uv run --project "$(pwd)" "$(pwd)/main.py"--project only sets the dependency-resolution root, not the working
directory uv run resolves the script argument against, so the script
path needs to be absolute.
--scope user makes it available in every project on this machine, since
ragd itself already scopes results per-repo. Verify with claude mcp list
or /mcp in a session.
Apache-2.0. See LICENSE.
Fuzzy/semantic RAG search over a codebase indexed by ragd.
Prefer this over Grep/Glob when the question is conceptual rather than an exact string/symbol match -- e.g. "where do we handle retry logic" or "how is auth configured" -- where you don't know the exact symbol/filename to search for. Returns a synthesized answer plus the source file excerpts it was drawn from.
Args: query: Natural-language question about the codebase. resource: Name of the ragd resource to search. If omitted, the resource whose indexed directory contains the current working directory is used automatically. top_k: Number of source chunks to retrieve and ground the answer in (default 5, max 20).
Inputs: query (required), top_k, resource. Returns: result
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Scored across 1 tool
With only one tool, there is no possibility of overlapping or confusable operations. query_codebase is clearly scoped to fuzzy/semantic codebase search rather than exact-match grep/glob.
query_codebase uses a clear verb_noun pattern and accurately describes the action and target. With a single tool, there is no naming inconsistency to create confusion.
One tool is the absolute minimum surface, making the server feel thin as a standalone integration. The narrow query-focused scope is defensible, but it is a borderline rather than fully developed toolset.
The server covers the core semantic search query flow, but it offers no way to list, inspect, or manage ragd resources. An agent can query successfully using auto-selection, yet resource discovery and index status information are notable gaps.
Search indexed code, trace dependencies, assess change impact, and recall repository memory.
Code intelligence for coding agents: semantic, AST, graph, and full-text search. 279+ languages.
Code intelligence for LLMs. Analyze, search, and retrieve code from any public git repository.
Ask a codebase what calls what: search, blast radius, paths between symbols, and diffs.
