OpenHands MCP Server

What is OpenHands MCP Server?

An MCP (Model Context Protocol) server that extends the capabilities of OpenHands. It provides three standalone tools—for running shell commands, fetching APIs, and simulating audio analysis—plus optional credential management and a Facebook Messenger bot.

How to use OpenHands MCP Server?

Clone the repository, make scripts executable (chmod +x run_lint.sh), verify Node.js 18+ and Python 3 are installed, then configure the three tools as stdio_servers in OpenHands (Settings → MCP) using the provided JSON snippet. Test each script individually with the given command-line examples.

Key features of OpenHands MCP Server

• analyze_audio.py – simulates audio transcription and summarization
• call_api.js – fetches JSON from any URL with comprehensive error handling
• run_lint.sh – simulates a lint check on a codebase
• No external dependencies beyond Python, Node.js, and Bash
• Optional password manager and .env to store API credentials
• Facebook bot for automated customer service (VIV Clinic)

Use cases of OpenHands MCP Server

• Fetch live data (e.g., Bitcoin price) and let OpenHands summarize it
• Run a simulated linter to check code quality inside OpenHands
• Transcribe a podcast episode using the audio analysis tool
• Deploy an automated Facebook Messenger bot that saves inquiries to Google Sheets

FAQ from OpenHands MCP Server

What does the OpenHands MCP Server do?

It adds three MCP tools to OpenHands: one runs shell commands, one calls external APIs, and one simulates audio processing. All communicate via stdio.

What runtime dependencies are required?

Python 3.x for analyze_audio.py, Node.js 18+ for call_api.js, and Bash for run_lint.sh. No extra packages are needed—all scripts use built‑in libraries.

How do I configure the server inside OpenHands?

In OpenHands go to Settings → MCP and add each script as a stdio_servers entry with its command and arguments (see the README’s JSON example).

Can I authenticate API requests?

Yes. You can store credentials in a .env file and use the included password_manager.py to retrieve them. The call_api_with_auth.js script supports bearer and Facebook authentication.

Are there any known limitations?

The audio and lint tools are simulations—they do not perform real transcription or linting. The server uses stdio transport only (no SSE). The VIV Clinic bot is separate from the three MCP tools.