neuralcontrol-mcp 1.3.0

MCP Server for the AI Control Plane — gives AI editors live access to endpoint configs, metrics, and SDK docs

neuralcontrol-mcp

MCP Server for the AI Control Plane — gives AI editors (Cursor, Claude, Windsurf) live access to your endpoint configs, real-time metrics, Request Coalescing examples, and SDK documentation.

Installation

pip install neuralcontrol-mcp

Configuration

Add this to your AI editor's MCP config (e.g. ~/.cursor/mcp.json or claude_desktop_config.json):

{
  "mcpServers": {
    "neuralcontrol": {
      "command": "neuralcontrol-mcp",
      "env": {
        "CONTROL_PLANE_URL": "https://api.neuralcontrol.online",
        "NEURALCONTROL_API_KEY": "acp_your_key_here"
      }
    }
  }
}

[!NOTE] Use https://api.neuralcontrol.online (Live) or http://localhost:8000 (Local).

Or copy .env.example to .env and fill in your values, then run directly:

neuralcontrol-mcp
# or
python -m server

What you get

Once connected, your AI editor can:

Tool	What it does
get_endpoint_config	Live AI flags (caching, circuit breaker, rate limits) for any endpoint
get_recent_metrics	Real-time p50/p95/p99 latency + error rates
get_adaptive_timeout_status	Which endpoints are currently timing out
get_ai_thresholds	Raw AI-calculated threshold values
get_all_ai_insights	Why the AI enabled/disabled each feature
get_open_incidents	Active latency spikes or error surges
get_active_overrides	Manual overrides currently in effect
create_threshold_override	Immediately override a threshold (e.g. during an incident)
get_sdk_setup_instructions	Get integration code for your service
get_feature_documentation	Full docs for any protection feature
get_feature_flags	Fetch all feature flags for a service
set_feature_flag	Create/update a feature flag's rollout
get_trace	Fetch full distributed trace waterfall

Environment Variables

Variable	Description	Default
CONTROL_PLANE_URL	URL of your running control-plane backend	http://localhost:8000
NEURALCONTROL_API_KEY	Your NeuralControl API key	(required)

Hosted / Docker

To run as a persistent HTTP/SSE server (for team-wide access):

docker build -t neuralcontrol-mcp .
docker run -p 8080:8080 \
  -e CONTROL_PLANE_URL=https://your-api.yourdomain.com \
  -e NEURALCONTROL_API_KEY=acp_xxx \
  neuralcontrol-mcp

Then connect via URL:

{ "mcpServers": { "neuralcontrol": { "url": "http://localhost:8080/sse" } } }

License

MIT