ragmap

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.

Category
访问服务器

README

RAGMap (RAG MCP Registry Finder)

Glama

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

RAGMap architecture diagram

<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 worker
  • apps/mcp-remote: Remote MCP server (Streamable HTTP)
  • packages/mcp-local: Local MCP server (stdio)
  • packages/shared: Zod schemas + shared types
  • docs: 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 (includes embeddings: true|false when semantic search is on/off)
  • GET /readyz
  • GET /v0.1/servers
  • GET /v0.1/servers/:serverName/versions
  • GET /v0.1/servers/:serverName/versions/:version (supports latest)
  • GET /rag/search
  • GET /rag/categories
  • GET /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 (stdio or streamable-http)
  • registryType (string)
  • hasRemote (true or false — 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 — optional discoveryService in server.json so clients can show “Discover more”
  • docs/AGENT-USAGE.mdfor agents: discovery, REST API, MCP install (no human intervention)
  • docs/DEPLOYMENT.md
  • docs/OVERVIEW.md
  • docs/DATA_MODEL.md
  • docs/PRIVACY.md
  • docs/PUBLISHING.md
  • docs/GLAMA-CHECKLIST.md
  • docs/GLAMA-DOCKERFILE.md
  • scripts/glama-score-status.sh — print public Glama score flags (inspectable/release/usage)

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选