MCP Interface Overview

TaskFlow exposes a Model Context Protocol (MCP) endpoint per project, allowing AI agents to call TaskFlow's task execution, status management, and knowledge-base capabilities directly.

Endpoint

POST /api/projects/{project_id}/mcp

Supports GET (availability probe), HEAD (health check), POST (message), DELETE.

Authentication

Use a project API key (created in Project Settings → API Keys):

Authorization: Bearer tf_b1af5033_<secret>

Key format: tf_{first-8-hex-of-project-id}_{32-char-secret}

Integrating with Claude Code

Add to .claude/mcp.json:

{
  "mcpServers": {
    "taskflow": {
      "type": "http",
      "url": "http://localhost:8000/api/projects/YOUR_PROJECT_ID/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Dynamic task tools

When a task type has Allow MCP enabled, it becomes an MCP tool named after its identifier. Calling it creates a task instance and returns immediately; poll with get_task.