ragmap
MapRag is a discovery + routing layer for retrieval. It indexes RAG-capable MCP servers, enriches them with structured metadata, and helps agents (and humans) quickly find the right retrieval server for a task under constraints like citations, freshness, privacy, domain, and latency. MapRag does not do RAG itself. It helps you choose the best RAG tool/server to do the retrieval.
README
RAGMap (RAG MCP Registry Finder)
RAGMap is a lightweight MCP Registry-compatible subregistry + MCP server focused on RAG-related MCP servers.
It:
- Ingests the official MCP Registry, enriches records for RAG use-cases, and serves a subregistry API.
- Exposes an MCP server (remote Streamable HTTP + local stdio) so agents can search/filter RAG MCP servers.
MapRag (RAGMap)
MapRag is a discovery + routing layer for retrieval. It helps agents and humans answer: which retrieval MCP server should I use for this task, given my constraints?
RAGMap does not do retrieval itself. It indexes and enriches retrieval-capable servers, then routes you to the right tool/server.
What you get after install (plain English)
- You get discovery/routing tools (
rag_find_servers,rag_get_server,rag_list_categories,rag_explain_score). - RAGMap helps you find the best retrieval server for your task and constraints.
- Your agent then connects to that chosen server to do the actual retrieval.
RAGMap does not:
- Ingest your private documents automatically.
- Host your personal vector database.
- Replace your end-to-end RAG pipeline.
If you need retrieval over your own data, use a retrieval server from RAGMap results (or your own server) that supports your ingest/index flow.
Freshness and ingest
- Hosted RAGMap updates its index on a schedule. Newly published/changed servers may appear with some delay.
- Most users do not run ingest themselves when using the hosted service.
- If you need tighter freshness control or private indexing behavior, self-host and run your own ingest schedule (
docs/DEPLOYMENT.md).
Features: Registry-compatible API; semantic + keyword search (when OPENAI_API_KEY is set, e.g. from env or your deployment’s secret manager); categories and ragScore; filter by hasRemote, reachable (HEAD-checked), citations, localOnly, transport, minScore, categories. Human browse UI at ragmap-api.web.app/browse — search, filter, copy Cursor/Claude config. MCP tools: rag_find_servers, rag_get_server, rag_list_categories, rag_explain_score.
Full overview: docs/OVERVIEW.md
Architecture

<details> <summary>Mermaid source</summary>
%%{init: {"theme":"base","themeVariables":{"primaryColor":"#ffffff","primaryTextColor":"#000000","primaryBorderColor":"#000000","lineColor":"#000000","secondaryColor":"#ffffff","tertiaryColor":"#ffffff","clusterBkg":"#ffffff","clusterBorder":"#000000","edgeLabelBackground":"#ffffff"},"flowchart":{"curve":"linear","nodeSpacing":75,"rankSpacing":70}}}%%
flowchart TB
%% Concept-only diagram (product value; no deployment/framework/datastore details)
classDef mono fill:#ffffff,stroke:#000000,color:#000000,stroke-width:1px;
subgraph Inputs[" "]
direction LR
subgraph Query["Agent-native interface"]
direction TB
Users["Agents + humans"]:::mono
subgraph Tooling["Tool call"]
direction LR
Criteria["Routing constraints<br/>domain, privacy, citations,<br/>freshness, auth, limits"]:::mono
Tools["MCP tools<br/>rag_find_servers<br/>rag_get_server<br/>rag_list_categories<br/>rag_explain_score"]:::mono
end
Users --> Criteria --> Tools
end
subgraph Subregistry["Subregistry (read-only)"]
direction TB
subgraph Ingest["Ingest"]
direction LR
Sources["Upstream MCP registries<br/>(official + optional)"]:::mono
Sync["Sync + normalize<br/>(stable schema)"]:::mono
Catalog["Enriched catalog<br/>(servers + versions)"]:::mono
Sources --> Sync --> Catalog
end
subgraph Enrich["Enrich (adds value)"]
direction LR
Cap["Structured metadata<br/>domain: docs|code|web|mixed<br/>retrieval: dense|sparse|hybrid (+rerank)<br/>freshness: static|continuous (max lag)<br/>grounding: citations|provenance<br/>privacy/auth: local|remote + req|optional<br/>limits: top_k|rate|max ctx"]:::mono
Trust["Trust signals (lightweight)<br/>status, reachability,<br/>schema stability, reports"]:::mono
end
Catalog --> Cap
Catalog --> Trust
end
end
subgraph Selection["Selection (the added value)"]
direction LR
Router["Router<br/>match + rank + explain"]:::mono
Ranked["Ranked candidates<br/>+ reasons + connect info"]:::mono
Retrieval["Chosen retrieval MCP server(s)<br/>(do retrieval)"]:::mono
Router --> Ranked --> Retrieval
end
Tools --> Router
Catalog --> Router
%% Keep the layout without adding a third visible "box" around Inputs.
style Inputs fill:#ffffff,stroke:#ffffff,stroke-width:0px
</details>
Monorepo layout
apps/api: REST API + MCP registry-compatible endpoints + ingestion workerapps/mcp-remote: Remote MCP server (Streamable HTTP)packages/mcp-local: Local MCP server (stdio)packages/shared: Zod schemas + shared typesdocs: docs + Firebase Hosting static assets
Local dev
cp .env.example .env
corepack enable
pnpm -r install
pnpm -r dev
Optional: set OPENAI_API_KEY in .env (see .env.example) to enable semantic search locally; GET /health will show "embeddings": true.
API: http://localhost:3000
MCP remote: http://localhost:4000/mcp
Ingest
curl -X POST http://localhost:3000/internal/ingest/run \
-H "Content-Type: application/json" \
-H "X-Ingest-Token: $INGEST_TOKEN" \
-d '{"mode":"full"}'
MCP usage
Remote (Streamable HTTP):
claude mcp add --transport http ragmap https://<your-mcp-domain>/mcp
Local (stdio, npm):
npx -y @khalidsaidi/ragmap-mcp@latest
Local (stdio):
pnpm -C packages/mcp-local dev
Key endpoints
GET /embed— embeddable “Search RAG MCP servers” widget (iframe; query params:q,limit)GET /health(includesembeddings: true|falsewhen semantic search is on/off)GET /readyzGET /v0.1/serversGET /v0.1/servers/:serverName/versionsGET /v0.1/servers/:serverName/versions/:version(supportslatest)GET /rag/searchGET /rag/categoriesGET /api/stats(public usage aggregates; no PII)GET /api/usage-graph(HTML chart of usage)POST /internal/ingest/run(protected)
For hosted ragmap-api.web.app, /internal/* routes are not exposed publicly.
GET /rag/search query params:
q(string)categories(comma-separated)minScore(0-100)transport(stdioorstreamable-http)registryType(string)hasRemote(trueorfalse— only servers with a remote endpoint)reachable(true— only servers whose streamable-http URL passed a HEAD check)citations(true— only servers that mention citations/grounding in metadata)localOnly(true— only stdio, no remote)
Smoke tests
API_BASE_URL=https://ragmap-api.web.app ./scripts/smoke-public.sh
MCP_URL=https://ragmap-api.web.app/mcp ./scripts/smoke-mcp.sh
Docs
docs/DISCOVERY-LINK-CONVENTION.md— optionaldiscoveryServicein server.json so clients can show “Discover more”docs/AGENT-USAGE.md— for agents: discovery, REST API, MCP install (no human intervention)docs/DEPLOYMENT.mddocs/OVERVIEW.mddocs/DATA_MODEL.mddocs/PRIVACY.mddocs/PUBLISHING.mddocs/GLAMA-CHECKLIST.mddocs/GLAMA-DOCKERFILE.mdscripts/glama-score-status.sh— print public Glama score flags (inspectable/release/usage)
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.