agent-runtime-control-tower

> FastAPI control tower for governing AI agents, MCP servers, runtime policies, approvals, replayable runs, and runtime state.

[](https://www.python.org/) [](https://fastapi.tiangolo.com/) [](#)

Everyone wants AI agents in production. Very few teams have a control layer for what those agents are allowed to do.

Agent Runtime Control Tower is a backend-first platform primitive for enterprise AI systems. It models the runtime governance layer around agents and MCP servers:

• which agents are allowed to run
• which MCP servers they can access
• when human approval is required
• how risky runs are blocked or escalated
• how tool calls are traced
• how incidents are created for unsafe behavior
• how runtime state can be stored in Redis
• how execution history can be replayed under tighter controls

• Python backend architecture for AI platform primitives
• MCP-aware runtime governance instead of uncontrolled tool access
• deterministic policy enforcement around agent execution
• tenant-aware access control and approval routing
• execution traces and incident visibility for AI operations
• DATABASE_URL-driven persistence ready for SQLite or PostgreSQL
• Redis-friendly runtime state for fast operational introspection

• GET /health
• GET /me
• GET /organizations
• GET /mcp-servers
• GET /agents
• GET /policies
• POST /runs
• GET /runs
• GET /runs/{run_id}/traces
• GET /runs/{run_id}/state
• POST /runs/{run_id}/replay
• GET /approvals
• POST /approvals/{approval_id}/decision
• GET /incidents

python -m venv .venv
.venv