0nmcp-crewai

An MCP server for CrewAI Enterprise — drive crew runs from any MCP client. Built by 0nORK.

What it does

Wraps the per-crew CrewAI Enterprise REST API (https://<crew>.crewai.com) as a standard Model Context Protocol server. Any MCP client — Claude Code, Cursor, Windsurf, 0nMCP, etc. — can now kick off, monitor, and resume CrewAI crew runs as first-class tools.

| Tool | Endpoint | Purpose | |---|---|---| | crewai_kickoff | POST /kickoff | Start a crew run with the given inputs | | crewai_status | GET /status/{kickoff_id} | Check progress of a running crew | | crewai_get_inputs | GET /inputs | List the inputs a crew expects | | crewai_resume | POST /resume | Resume a paused crew waiting on human input |

Free tier

CrewAI Enterprise gives every account 50 crew runs/month for free. This MCP server makes those 50 runs callable from anywhere AI runs — without writing CrewAI-specific code.

Install

Claude Desktop / Claude Code

Add to your MCP config:

{
  "mcpServers": {
    "crewai": {
      "command": "npx",
      "args": ["-y", "0nmcp-crewai"],
      "env": {
        "CREWAI_BEARER_TOKEN": "your_crewai_token"
      }
    }
  }
}

Cursor / Windsurf

Same pattern — point at npx -y 0nmcp-crewai with CREWAI_BEARER_TOKEN in env.

0nMCP / 0nCore

Connect from 0ncore.com/dashboard/admin/mcp-registry once this server is published to the official MCP registry. (Until then, use the CLI install above.)

Usage

Kick off a crew

// Tool call:
{
  "name": "crewai_kickoff",
  "arguments": {
    "crew_name": "research-crew",
    "inputs": {
      "topic": "AI agent frameworks",
      "depth": "deep"
    }
  }
}

// Returns:
{ "kickoff_id": "abc123-..." }

Check status

{
  "name": "crewai_status",
  "arguments": {
    "crew_name": "research-crew",
    "kickoff_id": "abc123-..."
  }
}

List a crew's inputs

{
  "name": "crewai_get_inputs",
  "arguments": { "crew_name": "research-crew" }
}

Resume a paused run

{
  "name": "crewai_resume",
  "arguments": {
    "crew_name": "research-crew",
    "kickoff_id": "abc123-...",
    "task_outputs": { "human_response": "approved" }
  }
}

Crew identification

You can identify a crew two ways:

| Form | Example | Resolves to | |---|---|---| | crew_name | "research-crew" | https://research-crew.crewai.com | | crew_url | "https://research-crew.crewai.com" | (passed through verbatim) |

A default can also be set globally with CREWAI_DEFAULT_CREW_URL.

Auth

| Method | Where | Priority | |---|---|---| | Per-call | bearer argument on any tool | Highest | | Global env | CREWAI_BEARER_TOKEN | Fallback |

Why a separate MCP package vs. just a 0nMCP service?

This package implements the Model Context Protocol directly so it works in any MCP client — not just 0nMCP. If you only use 0nMCP, the CrewAI tools are also bundled into 0nMCP's first-party catalog (crewai_*) for zero-config use. This standalone package exists for the broader MCP ecosystem.

License

MIT — © 2026 RocketOpp LLC.

Related

• 0nMCP — Universal AI API Orchestrator (1,594 tools across 105 services)
• Model Context Protocol — the protocol this server implements
• CrewAI — the agent platform this wraps