Cadence MCP Server (Model Context Protocol)

Official Model Context Protocol (MCP) server for the Cadence Meeting & Speech Intelligence Platform.

Connects autonomous AI coding agents (Cursor, Claude Desktop, Claude Code, Windsurf, Copilot) directly to your team's live meeting context, verbal directives, anti-patterns, and architectural decisions.

⚡ What Cadence MCP Does

1. Organizational Decision Verification:
   Ask: "Hey Claude, did we decide to go with MongoDB or Postgres in our architecture sync?"
   Claude invokes search_cadence_knowledge to retrieve the exact spoken decision and quotes.
2. Post-Meeting Code Handoff:
   Ask: "Implement the cache invalidation logic we just finished discussing."
   Claude invokes get_latest_meeting to extract verified verbal directives and implementation rules.
3. Meeting Discovery:
   Query meetings by date or keyword (e.g. list_meetings(query="sprint")) to pull historical architectural constraints.
4. Anti-Pattern Guardrails:
   Automatically loads explicit architectural rejections (e.g. "Do not introduce MongoDB into the relational service") so AI agents don't write disallowed code.

🛠️ MCP Tools Exposed

🚀 Setup & IDE Configuration

Step 1: Generate Your API Key

In your Cadence web app, navigate to Profile → Developer / API Keys and click Generate API Key. Copy your cad_live_... key.

Method A: Option 1 — Remote Cloud SSE (Recommended, Zero Local Installation)

No local Python environment or packages required! Your IDE connects directly to the Cadence API Gateway over Server-Sent Events (SSE).

1. Cursor IDE / Windsurf

Add to .cursor/mcp.json in your workspace root, or configure via Cursor Settings → Features → MCP:

{
  "mcpServers": {
    "cadence": {
      "url": "http://localhost:8000/mcp/sse",
      "headers": {
        "Authorization": "Bearer cad_live_your_api_key_here"
      }
    }
  }
}

(In production, replace http://localhost:8000/mcp/sse with your production URL, e.g. https://api.yourcadenceapp.com/mcp/sse)

2. Claude Desktop (SSE Support)

Add to your claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "cadence": {
      "url": "http://localhost:8000/mcp/sse",
      "headers": {
        "Authorization": "Bearer cad_live_your_api_key_here"
      }
    }
  }
}

Method B: Option 2 — Local stdio CLI Runner

If you prefer running the MCP server locally via standard I/O (e.g. for air-gapped environments or local package development):

1. Cursor IDE (.cursor/mcp.json)

{
  "mcpServers": {
    "cadence": {
      "command": "python",
      "args": ["-m", "cadence_mcp"],
      "env": {
        "CADENCE_API_KEY": "cad_live_your_api_key_here",
        "CADENCE_BASE_URL": "http://localhost:8000"
      }
    }
  }
}

Or with uvx (zero installation CLI runner):

{
  "mcpServers": {
    "cadence": {
      "command": "uvx",
      "args": ["cadence-mcp"],
      "env": {
        "CADENCE_API_KEY": "cad_live_your_api_key_here",
        "CADENCE_BASE_URL": "http://localhost:8000"
      }
    }
  }
}

2. Claude Code (CLI)

claude mcp add cadence -e CADENCE_API_KEY=cad_live_... -e CADENCE_BASE_URL=http://localhost:8000 -- python -m cadence_mcp

🔒 Security & Multi-Tenant Isolation

• Transport Security: Credentials (CADENCE_API_KEY) are handled strictly by the local MCP server over HTTP Authorization headers—the LLM never passes credentials in prompt context.
• Tenant Isolation: Every query in the Cadence backend is strictly constrained to meetings where user_id == meeting_participant.user_id. Calling another organization's meeting_id returns an immediate 403 Forbidden.