OpenHands

OpenHands is an open-source autonomous coding agent. This guide walks through connecting OpenHands to Loop so it can pull work from the dispatch queue, execute tasks, and report results back.

OpenHands integrates with Loop through two surfaces:

• Microagent -- a keyword-triggered markdown file that teaches OpenHands how to use the Loop API.
• MCP Server (optional) -- provides tool-based access to Loop for richer agent interactions.

Prerequisites

• A running Loop API instance (local or hosted)
• A valid LOOP_API_KEY (authentication docs)
• OpenHands installed and configured (OpenHands docs)

Microagent Setup

OpenHands supports microagents -- small markdown files that are loaded into the agent's context when specific keywords appear in the conversation. Loop provides a pre-built microagent that teaches OpenHands the full Loop API.

mkdir -p .openhands/microagents

npx @dork-labs/loop-skill init --openhands

---
name: loop
agent: CodeActAgent
trigger_type: keyword
triggers:
  - loop
  - looped
  - looped.me
  - loop api
  - loop issue
  - ingest signal
---

export LOOP_API_KEY=loop_your_api_key_here

Check the Loop API health endpoint and list any issues in triage status.

Microagent Content

The Loop microagent provides OpenHands with:

• API base URL and authentication pattern (Authorization: Bearer $LOOP_API_KEY)
• Key endpoints -- issues, signals, projects, goals, dashboard, and dispatch
• Code examples for creating issues and ingesting signals
• Links to full documentation and the machine-readable llms.txt index

When the microagent is active, OpenHands can create issues, ingest signals, check the dispatch queue, and report results without any additional configuration.

Adding MCP Support

For richer tool-based interaction, you can also configure the Loop MCP server. This gives OpenHands named tools like loop_get_next_task and loop_create_issue instead of relying on raw HTTP calls.

If your OpenHands setup supports MCP, add the Loop server to your MCP configuration:

{
  "mcpServers": {
    "loop": {
      "command": "npx",
      "args": ["-y", "@dork-labs/loop-mcp"],
      "env": {
        "LOOP_API_KEY": "<your-api-key>",
        "LOOP_API_URL": "http://localhost:5667"
      }
    }
  }
}

See the MCP Server docs for the full list of available tools and configuration options.

Autonomous Polling Pattern

To run OpenHands in a continuous loop -- pulling work from Loop, executing it, and reporting back -- use this pattern in your OpenHands prompt or automation script:

1. Poll for work -- call GET /api/dispatch/next to claim the highest-priority unblocked issue.
2. Execute the task -- use the returned prompt field as instructions. If prompt is null, use the issue title and description.
3. Report results -- add a comment to the issue with POST /api/issues/:id/comments describing what was done.
4. Update status -- mark the issue as done with PATCH /api/issues/:id and set status: "done".
5. Repeat -- poll for the next issue.

# Example: poll for work and show what's next
curl -s http://localhost:5667/api/dispatch/next \
  -H "Authorization: Bearer $LOOP_API_KEY" | jq .

Next Steps

• Dispatch -- understand how Loop selects and scores work items.
• Signals -- learn how to feed external data into the Loop triage queue.
• Agent Skill -- explore the full skill package for additional agent context.
• MCP Server -- browse the complete list of MCP tools.