Ragmap

What khalidsaidi/ragmap MCP server does

The khalidsaidi/ragmap MCP server is a discovery and routing layer for retrieval-oriented MCP servers. It maintains a focused subregistry based on the official MCP Registry, adds metadata useful for RAG selection, and exposes search and inspection interfaces for agents and developers.

The project does not retrieve documents, manage a private vector database, or ingest a user's files. Instead, it returns candidate servers and connection information so an agent can connect to a selected retrieval server for the actual search or generation workflow.

A hosted service is available at https://ragmap-api.web.app, including a browser interface and MCP endpoint. The repository also supports local development and a local stdio MCP package.

How it works

Registry records are normalized and enriched with fields such as domain coverage, retrieval approach, freshness behavior, grounding or citation support, privacy and authentication characteristics, transport, and usage limits. Lightweight trust signals include service status, recent reachability checks, schema stability, and reports.

Agents can search or filter the catalog by query, categories, minimum score, transport, registry type, remote availability, recent reachability, citation metadata, or local-only operation. The service can combine keyword search with semantic search when OPENAI_API_KEY is configured. Results include ranking information, and rag_explain_score can provide reasons behind a score.

The MCP interface exposes four tools: rag_find_servers, rag_get_server, rag_list_categories, and rag_explain_score. The REST API provides related endpoints for search, top recommendations, installation details, categories, statistics, health, and registry-compatible server records.

Setup and configuration

For a hosted connection, the README gives this Claude MCP command:

claude mcp add --transport http ragmap https://<your-mcp-domain>/mcp

The hosted deployment can be queried directly with HTTP. For example, /rag/search supports filters for recent reachability and remote availability, while /rag/install returns installation configuration for a selected server.

To run the local MCP package over stdio, use:

npx -y @khalidsaidi/ragmap-mcp@latest

Local repository development uses corepack enable, pnpm -r install, and pnpm -r dev. The API runs on port 3000 and the local remote MCP service on port 4000. Semantic search is optional; setting OPENAI_API_KEY enables it locally. Self-hosted ingestion uses a protected X-Ingest-Token and is intended for operators managing their own index schedule.

Tools and capabilities

• Search for retrieval MCP servers using natural-language queries.
• Filter candidates by categories, score, transport, citations, remote availability, local-only status, and recent reachability.
• Retrieve a server record and installation configuration.
• List available RAG categories.
• Inspect why a candidate received its ranking.
• Browse and copy Cursor or Claude configuration from the hosted UI.

The khalidsaidi/ragmap MCP server is useful for routing decisions where backend choice depends on constraints such as privacy, citations, freshness, or deployment transport.

Limitations and notes

Index updates occur on a schedule in the hosted deployment, so new or changed servers may not appear immediately. Reachability is based on probe checks for streamable HTTP or SSE endpoints and may become stale. Semantic search depends on an OpenAI API key; without it, keyword search remains available.

The khalidsaidi/ragmap MCP server does not replace an end-to-end RAG pipeline. It does not automatically ingest private documents, host a user's vector store, or perform retrieval on behalf of the caller. Users must connect to a returned retrieval server, or provide another server that supports their ingestion and indexing process.