Local RAG tools with MCP support
MCP (Model Context Protocol) is an open standard, first released by Anthropic, for plugging tools and data into AI apps. Claude Desktop, ChatGPT, Cursor, VS Code and many others can connect to MCP servers. For RAGShelf readers that means one thing: your documents stay indexed on your machine, and the AI app you already use can search them. (modelcontextprotocol.io)
MCP server
The tool exposes your indexed documents. Point Claude Desktop or Cursor at it and they can search your files. This is the one you want for "chat with my local docs from Claude".
MCP client
The tool uses other MCP servers (web search, GitHub, a database) inside its own chat or agents. Handy, but it doesn't share your docs with other apps.
How to connect local docs to Claude Desktop, Cursor or ChatGPT
- Pick a tool below marked Server and index your documents in it.
- Start its MCP server (each source link shows the exact command or URL).
- Add the server to your app: in Claude Desktop, edit its MCP config (Settings → Developer); in Cursor, add it under MCP settings; ChatGPT connects to remote MCP servers through connectors.
- Ask a question. The app calls the tool, which searches your local index and returns passages.
Heads up: with a cloud app like Claude or ChatGPT, the passages it retrieves are sent to that cloud model. Your full library stays local, but the snippets used to answer don't. For fully offline use, pair an MCP client tool with a local model (see hybrid setups).
Tools with MCP support (24)
| Tool | MCP | What it does | Source |
|---|---|---|---|
| AnythingLLM | MCP: Client | Connects to MCP servers so its agents can use outside tools. | source |
| Cherry Studio | MCP: Client | Adds MCP servers to extend the desktop app. | source |
| Dify | MCP: Server + client | Publish a Dify app as an MCP server; use MCP tools in apps. | source |
| Docling | MCP: Server | Official Docling MCP server for document conversion. | source |
| DocsGPT | MCP: Server + client | MCP server lets Claude, Cursor or any MCP client search your agent's knowledge. | source |
| Flowise | MCP: Client | MCP tool nodes in the flow builder. | source |
| Haystack | MCP: Server (via Hayhooks) | Serve Haystack pipelines as MCP servers with the official Hayhooks project. | source |
| Jan | MCP: Client | MCP integration for agentic features. | source |
| KoboldCpp | MCP: Supported | README lists MCP server support for tools and agents. | source |
| Langflow | MCP: Server + client | Deploy flows as an MCP server for MCP clients. | source |
| LibreChat | MCP: Client | Uses MCP servers as tools for agents. | source |
| LlamaIndex | MCP: Client + server tools | Official llama-index-tools-mcp package. | source |
| LocalAI | MCP: Client | MCP support for its built-in agents. | source |
| MaxKB | MCP: Client | Workflow engine supports MCP tool use. | source |
| Msty Studio | MCP: Client | Toolbox supports MCP tools. | source |
| Onyx | MCP: Server + client | Point Claude Code, Codex or any MCP client at Onyx; agents can call MCP actions. | source |
| Open Notebook | MCP: Server | Connect Claude Desktop, VS Code and other MCP clients. | source |
| Open WebUI | MCP: Client | Connects to MCP servers (native or through the MCPO proxy). | source |
| PipesHub | MCP: Server | Search your connected sources over MCP (@pipeshub-ai/mcp). | source |
| PrivateGPT | MCP: Client | README lists MCP connectors and remote MCP servers for agents. | source |
| Quivr | MCP: Server | `quivr mcp` lets AI agents search and cite your Quivr knowledge. | source |
| R2R | MCP: Server | MCP server module in the R2R Python package. | source |
| RAGFlow | MCP: Server + client | Ships an MCP server for its retrieval and can connect external MCP servers. | source |
| SurfSense | MCP: Client | Adds third-party MCP servers from its plugin registry or by URL. | source |
We only tag a tool when its own README or docs say it supports MCP. Checked 2026-10-27. No MCP support found yet in official docs for: LightRAG, LocalGPT, Kotaemon, Khoj, GPT4All (LocalDocs), h2oGPT, Microsoft GraphRAG, Unstructured (OSS), Verba, Paperless-ngx, paperless-gpt, Ollama, RAG Web UI, AirLLM. Spot one we missed? Tell us.
The RAGShelf weekly
New local RAG tools, releases, and tested setups. One email a week, no spam.
Updates on X: @mkjerman