colabhive-mcp

Model Context Protocol (MCP) server that exposes ColabHive models — LLMs, specialists, tools and your trained models — to MCP-compatible clients such as Claude Desktop, Claude Code, Cursor and n8n.

User docs: https://docs.colabhive.com/mcp

Install

# Recommended (no global Python needed)
uvx colabhive-mcp

# Or pipx
pipx install colabhive-mcp

# Or pip (Python 3.10+)
pip install colabhive-mcp

Quickstart

Create an API key in the ColabHive Console (Settings → API Keys); keys start with hive_.

export COLABHIVE_API_KEY=hive_...
colabhive-mcp test

Then add the server to your MCP client:

{
  "mcpServers": {
    "colabhive": {
      "command": "uvx",
      "args": ["colabhive-mcp"],
      "env": { "COLABHIVE_API_KEY": "hive_..." }
    }
  }
}

A hosted server is also available at https://mcp.colabhive.com. See the user docs for per-client guides.

colabhive-mcp sdk-stdio and colabhive-mcp sdk-serve provide strict stdio and stateless Streamable HTTP transports built on the official MCP Python SDK, alongside the default stdio and HTTP/SSE transports.

Built-in account tools

• get_account_preferences and set_share_hive_opt_in read and set the account-scoped Share Hive preference. The credential selects the account; only the account owner can write it.
• get_effective_data_protection and preflight_data_protection read the effective node-eligibility policy and preflight a placement. There is no policy-mutation tool.

Experimental Cohort tools

Six opt-in tools (cohort_capabilities, plan_cohort, run_cohort, get_cohort, explain_cohort, cancel_cohort) stay hidden unless COLABHIVE_MCP_COHORT_TOOLS_ENABLED=true, MCP stability permits experimental tools, the account has Cohort capability, and the kind and side-effect filters allow the operation.

License

MIT — see LICENSE.

Links

• Docs: https://docs.colabhive.com/mcp
• Security: security@colabhive.com