@ordo/mcp-server

MCP (Model Context Protocol) Server for Ordo — Control Plane for AI Agents.

This server enables AI agents (like Claude in Cursor) to integrate with Ordo for:

• Activity Tracking — Track decisions, LLM calls, tool usage
• Cost Monitoring — Monitor token usage and spending
• Trust Scoring — Build trust through consistent behavior
• Policy Enforcement — Respond to governance rules

Installation

Using npx (Recommended)

No installation needed:

ORDO_API_KEY=your_key npx @ordo.space/mcp-server

Global Install

npm install -g @ordo.space/mcp-server

Then run:

ORDO_API_KEY=your_key ordo-mcp

Setup in Cursor

Add to your .cursor/mcp.json:

{
  "mcpServers": {
    "ordo": {
      "command": "npx",
      "args": ["-y", "@ordo.space/mcp-server"],
      "env": {
        "ORDO_API_URL": "https://api.ordo.ai",
        "ORDO_API_KEY": "your_api_key_here",
        "ORDO_AGENT_NAME": "cursor-agent"
      }
    }
  }
}

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | ORDO_API_KEY | Yes | - | Your Ordo API key (get from Settings > API Keys) | | ORDO_API_URL | No | http://localhost:3333 | Ordo API URL | | ORDO_AGENT_NAME | No | cursor-agent | Name for this agent | | ORDO_AUTO_TRACKING | No | true | Enable automatic tool call tracking |

Auto-Tracking

Starting from v0.3.0, the MCP server includes automatic telemetry that tracks all tool calls without requiring explicit tracking calls.

How It Works

┌─────────────────────────────────────────────────────────────────┐
│                      MCP Server                                 │
├─────────────────────────────────────────────────────────────────┤
│  Level 1: AUTO-TELEMETRY (always on by default)                │
│  • Tracks ALL tool calls automatically                          │
│  • Minimal data: tool_name, timestamp, duration, success       │
│  • Source: "auto" — separate from explicit tracking            │
│  • Fire-and-forget — never blocks agent execution              │
├─────────────────────────────────────────────────────────────────┤
│  Level 2: SEMANTIC TRACKING (explicit calls)                   │
│  • Agent calls ordo_track_* tools                               │
│  • Rich data: reasoning, confidence, context                   │
│  • Source: "explicit" — higher detail                          │
│  • Recommended for important decisions                          │
└─────────────────────────────────────────────────────────────────┘

Benefits

• Never miss data: Even if agent forgets to track, baseline telemetry is captured
• No performance impact: Tracking is async and non-blocking
• No conflicts: Auto-tracking and explicit tracking complement each other
• Opt-out: Set ORDO_AUTO_TRACKING=false to disable

Excluded Tools

Auto-tracking ignores Ordo tracking tools themselves to avoid infinite loops:

• ordo_start_task, ordo_track_decision, ordo_track_llm_call
• ordo_track_tool_call, ordo_track_error, ordo_complete_task
• ordo_check_status

Available Tools

Task Management

ordo_start_task

Start tracking a new task or workflow.

{
  "task_id": "pr-review-123",
  "task_name": "Review Pull Request",
  "subject_type": "pull_request",
  "subject_id": "PR-123"
}

ordo_complete_task

Mark a task as complete.

{
  "task_id": "pr-review-123",
  "status": "completed",
  "summary": "Approved with minor suggestions"
}

Event Tracking

ordo_track_decision

Track a decision made by the agent.

{
  "action": "approve",
  "confidence": 0.95,
  "reasoning": "Code follows best practices",
  "task_id": "pr-review-123"
}

ordo_track_llm_call

Track an LLM API call for cost monitoring.

{
  "model": "claude-3.5-sonnet",
  "tokens": 1500,
  "latency_ms": 2100
}

ordo_track_tool_call

Track a tool or function call.

{
  "tool_name": "github_api",
  "tool_input": { "action": "get_pr" },
  "success": true,
  "latency_ms": 350
}

ordo_track_error

Track an error that occurred.

{
  "error_type": "timeout",
  "error_message": "API request timed out",
  "recoverable": true
}

Status Check

ordo_check_status

Check if the agent should continue operating.

{}

Returns:

{
  "status": "active",
  "trust_score": 0.95,
  "should_continue": true,
  "recommendation": "Agent operating normally."
}

Available Resources

Resources can be read to get Ordo data:

| Resource | Description | |----------|-------------| | ordo://agents | List of active agents | | ordo://policies | Active policies | | ordo://subjects | Subjects being processed | | ordo://status | System status | | ordo://agents/{id} | Agent details | | ordo://policies/{id} | Policy details | | ordo://subjects/{id} | Subject details |

Example Usage Flow

1. Start a task when beginning work:

   ordo_start_task(task_id="review-123", subject_type="pull_request", subject_id="PR-123")

2. Track decisions as you work:

   ordo_track_decision(action="approve", confidence=0.9, reasoning="Clean code")

3. Track LLM calls for cost monitoring:

   ordo_track_llm_call(model="claude-3.5-sonnet", tokens=1200)

4. Check status periodically:

   ordo_check_status()

5. Complete the task when done:

   ordo_complete_task(task_id="review-123", status="completed")

Self-Hosting

If running Ordo locally:

{
  "env": {
    "ORDO_API_URL": "http://localhost:3333",
    "ORDO_API_KEY": "your_local_api_key"
  }
}

Troubleshooting

"ORDO_API_KEY not set"

Make sure the API key is provided in the environment.

"Agent should STOP operations"

Your agent has been paused by a policy or human. Check the Ordo dashboard.

"Trust score is low"

Recent actions have lowered trust. Review the Audit Trail in Ordo.

License

MIT © Ordo Team