Mission Control

Mission Control is a real-time observability and coordination platform for autonomous AI agents. It provides a central dashboard where you can monitor every agent that is running, watch task progress events stream in live, track LLM token costs, and inspect artifacts — all without touching the agent code beyond a single SDK import.

---

Quick Start

Prerequisites

• Docker and Docker Compose

1. Clone and configure

git clone <repo-url>
cd mission_control
cp infra/.env.example infra/.env   # edit API keys and secrets

2. Start the full stack

docker compose -f infra/docker-compose.yml up --build

Services:

Service	URL
Dashboard (frontend)	http://localhost:3000
API (backend)	http://localhost:8000
Nginx proxy	http://localhost:80
PostgreSQL	localhost:5432
Redis	localhost:6379

3. Open the dashboard

Navigate to http://localhost in your browser.

---

Architecture Overview

Agents ──HTTP──► FastAPI backend ──► PostgreSQL (persistence)
                       │
                       ├──► Redis pub/sub
                       │         │
                       └──WS──► Next.js dashboard (live updates)
                            │
                         Nginx :80

• Backend (FastAPI + SQLAlchemy) — REST ingest API, WebSocket fan-out, event storage.
• Frontend (Next.js 15 + Tailwind) — live dashboard with agent cards, task timelines, and cost charts.
• PostgreSQL — primary persistent store for all agents, tasks, events, runs, and artifacts.
• Redis — pub/sub broker that synchronises WebSocket state across backend instances.
• Nginx — reverse proxy that routes /api/* and /ws/* to the backend and everything else to the frontend.

See docs/architecture.md for the full design document.

---

Python SDK

Install:

pip install mission-control-client
# or from source:
pip install -e ./agent-sdk/python

Basic usage:

import asyncio
from mission_control_client import MissionControlClient

async def main():
    async with MissionControlClient(
        base_url="http://localhost:8000",
        api_key="your-api-key",
        agent_key="my-agent-01",
        name="My Research Agent",
        runtime_type="custom",
        role="researcher",
        model_provider="anthropic",
        model_name="claude-sonnet-4-6",
    ) as mc:
        task = await mc.create_task("task-001", "Research NBA slate")
        await mc.assign_task(task["id"])
        await mc.task_started(task["id"])
        await mc.update_progress(task["id"], 50, "Fetching odds")
        await mc.complete_task(task["id"], "Done!")

asyncio.run(main())

See agent-sdk/python/examples/basic_usage.py for a complete working example covering manual task lifecycle, the @mc_task decorator, error handling, and cost tracking.

OpenClaw Wrapper

Wrap any subprocess (including OpenClaw jobs) without modifying it:

mc-openclaw \
    --api-key    your-api-key \
    --agent-key  openclaw-worker-01 \
    --task-key   "nba-2026-03-11" \
    --task-title "NBA Research" \
    -- python my_job.py --date 2026-03-11

The job can emit structured events by printing MC_EVENT: prefixed JSON:

import json
def mc_event(type_, message=None, **payload):
    obj = {"type": type_}
    if message: obj["message"] = message
    if payload: obj["payload"] = payload
    print("MC_EVENT:" + json.dumps(obj), flush=True)

mc_event("progress_updated", "Fetching odds", percent=45)
mc_event("tool_called", tool_name="web_search", args={"q": "NBA tonight"})

See docs/agent_integration.md for the full integration guide.

---

Documentation

Document	Description
docs/architecture.md	System design, component breakdown, scaling
docs/event_schema.md	All event types, payloads, lifecycle diagrams
docs/agent_integration.md	Python SDK, OpenClaw wrapper, raw HTTP, MC_EVENT protocol
docs/api_spec.md	Full REST and WebSocket API reference

---

Dashboard

┌─────────────────────────────────────────────────────────────────┐
│  Mission Control                                    ● 3 online  │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  Agents                                                          │
│  ┌──────────────────────┐  ┌──────────────────────┐            │
│  │ ● My Research Agent  │  │ ● OpenClaw Worker 01 │            │
│  │   researcher         │  │   worker             │            │
│  │   claude-sonnet-4-6  │  │   claude-haiku-3-5   │            │
│  │   CPU 23%  MEM 512MB │  │   CPU 45%  MEM 1.2GB │            │
│  └──────────────────────┘  └──────────────────────┘            │
│                                                                  │
│  Active Tasks                                                    │
│  ┌─────────────────────────────────────────────────────────┐   │
│  │ NBA Research 2026-03-11        ████████░░  80%  running │   │
│  │ Fetching odds — 12:01:35                                 │   │
│  └─────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘

---

Project Structure

mission_control/
├── agent-sdk/
│   └── python/
│       ├── mission_control_client/
│       │   ├── __init__.py          # Package exports
│       │   ├── client.py            # MissionControlClient
│       │   ├── decorators.py        # @mc_task, @mc_tool
│       │   └── openclaw_wrapper.py  # CLI wrapper for subprocesses
│       ├── examples/
│       │   ├── basic_usage.py       # Full SDK walkthrough
│       │   └── openclaw_example.py  # MC_EVENT protocol demo
│       └── setup.py
├── backend/                         # FastAPI application
├── frontend/                        # Next.js dashboard
├── infra/
│   ├── docker-compose.yml
│   ├── nginx/nginx.conf
│   └── postgres/init.sql
└── docs/
    ├── architecture.md
    ├── event_schema.md
    ├── agent_integration.md
    └── api_spec.md

---

License

MIT