jarvisassistantux/loopsense

LoopSense is an open-source MCP server that closes the feedback loop for AI coding agents — giving them real-time visibility into CI results, deployments, test outcomes, and file system changes.
What it does
When an AI agent pushes code, runs tests, or triggers a deployment, LoopSense watches the downstream effects and surfaces them back to the agent. No more blind actions.
**Supported sources:**
- GitHub Actions CI runs (polling)
- Local processes (stdout/stderr capture, exit codes)
- File system changes (via chokidar)
- HTTP endpoints (polling, with status/body assertions)
- Incoming webhooks (lightweight HTTP server)
Requirements
- Node.js 18+
Installation
Or run directly with npx:
MCP Configuration
**Claude Code** (one-liner):
Or add manually to your claude_desktop_config.json (or equivalent MCP host config):
Or if installed globally:
Environment Variables
| Variable | Description |
|---|---|
| GITHUB_TOKEN | GitHub personal access token for CI polling |
Tools
watch_ci
Watch a GitHub Actions workflow run. Polls every 30 seconds and emits events on status changes.
watch_process
Spawn a local process and capture its output and exit code.
watch_file
Watch a file or directory for changes.
watch_url
Poll an HTTP endpoint and detect status or body changes.
watch_webhook
Start a local HTTP server to receive webhook payloads.
Configure your webhook sender to POST to http://localhost:9876.
check_consequences
Get events for a specific action or all recent events.
list_watches
List all active watchers.
cancel_watch
Stop a watcher by ID.
poll_events
Get events since a timestamp (fallback for clients without notification support).
Resources
LoopSense exposes two MCP resources that update reactively:
- loopsense://timeline/recent — last 100 events across all watches
- loopsense://watches/active — all currently active watches
- loopsense://consequences/{action_id} — events for a specific action
Usage Example
An agent workflow might look like:
1. Agent pushes code to a branch
2. Agent calls watch_ci with action_id: "my-pr-123"
3. LoopSense polls GitHub Actions every 30 seconds
4. When CI completes, agent calls check_consequences with action_id: "my-pr-123"