Skip to content

Integrations

Bee MCP Server

Connect Bee to Claude Desktop, Claude Code, Cursor, VS Code, Zed, Windsurf, or OpenCode via the Model Context Protocol — Anthropic’s open standard for LLM ↔ tool integration. Bee ships 8 tools — chat, code, security, research, plus documents (RAG) and memory — over a hosted gateway (no GPU, no model download), and works with the free tier.

stdio transport — supported
On the official MCP Registry · io.github.cuilabs/bee

Install in one step

Add Bee to Cursor with a single click, or paste the config below into any MCP client. Both use uvx, which fetches bee-sdk and runs the bee-mcp server for you. Then set your BEE_API_KEY (create one at account → API keys; the free Cell tier works).

VS Code: code --add-mcp '{"name":"bee","command":"uvx","args":["--from","bee-sdk@latest","bee-mcp"]}' — or search @mcp in the Extensions view.

Manual install

  1. 1. Install the server. The bee-mcp server ships with the Python SDK (zero extra deps):
    pip install bee-sdk     # provides the bee-mcp console script
    # or run with no install:  uvx --from bee-sdk@latest bee-mcp
  2. 2. Get a key. Create a bee_sk_ key at account → API keys and export it as BEE_API_KEY.
  3. 3. Wire your client. Add the config for your editor below.

Client configuration

Claude Desktop · Claude Code · Cursor · VS Code · Windsurf · Zed

Config path: the client's mcpServers config (claude_desktop_config.json, .cursor/mcp.json, VS Code settings, …)

{
  "mcpServers": {
    "bee": {
      "command": "uvx",
      "args": ["--from", "bee-sdk@latest", "bee-mcp"],
      "env": { "BEE_API_KEY": "bee_sk_..." }
    }
  }
}

OpenCode

Config path: ~/.config/opencode/opencode.json (or project opencode.json)

{
  "mcp": {
    "bee": {
      "type": "local",
      "command": ["uvx", "--from", "bee-sdk@latest", "bee-mcp"],
      "environment": { "BEE_API_KEY": "bee_sk_..." }
    }
  }
}

Already have bee-sdk installed? Replace "command": "uvx", "args": ["--from", "bee-sdk@latest", "bee-mcp"] with "command": "bee-mcp".

Tools exposed (up to 8)

Each tool routes through the appropriate domain LoRA adapter on the hosted gateway. If an adapter isn’t promoted for a domain yet, the tool still answers from the base model — without domain specialisation. Honest fallback, no silent failure.

ToolPurposeDomain adapter
bee_chatAsk Bee anything; optional domain specialist + model tierany domain (plan-limited)
bee_codeExplain, fix, refactor, or write tests — pick via `action`programming
bee_securityCode audit, threat model, authorised pentest assist, or smart-contract review — pick via `task`cybersecurity / blockchain
bee_researchQuantum-circuit design or paper critique — pick via `task`quantum / research
bee_documents_searchSearch your documents (RAG) — top matching chunksplans with RAG
bee_documents_addAdd a document to your knowledge baseplans with RAG
bee_memory_searchRecall facts from your personal memorymemory opt-in
bee_memory_addRemember a fact for future conversationsmemory opt-in

Resources

  • bee://status — gateway connectivity + auth status
  • bee://domains — full domain list

Prefer code over MCP?

We also publish first-party SDKs that talk to the same Bee API directly:

Listed on the official MCP Registry as io.github.cuilabs/bee. Community examples at cuilabs/bee-community. Spec at modelcontextprotocol.io. Need help wiring an editor we don’t list? contact support.