Remote OpenCode MCP Server

A Model Context Protocol (MCP) server that enables remote access to OpenCode AI coding agent. This allows MCP-compatible clients (Claude Desktop, Cursor, etc.) to leverage OpenCode's capabilities.

Features

• Remote OpenCode Integration: Connect to OpenCode instances running anywhere

• MCP Protocol: Full MCP server implementation with SSE transport

• 4 Core Tools:

  • Create new sessions

  • Send prompts with automatic timeout handling

  • Check session status and history

  • List all sessions

• Configurable Timeout: Adjustable task timeout via environment variable

• Authentication Support: Optional password protection

Related MCP server: Shellagent MCP Server

Requirements

• Python 3.12+

• OpenCode running in serve mode

OpenCode Serve Mode

This MCP server requires OpenCode to be running in serve mode with API enabled:

opencode --serve

Or with custom port and password:

opencode --serve --port 4096 --password your_password

Make sure OPENCODE_URL and OPENCODE_SERVER_PASSWORD in your .env match the serve command.

Installation

From Source

git clone https://github.com/Dirichelet/remote-opencode-mcp.git
cd remote-opencode-mcp
uv sync

Using uv

uv add remote-opencode-mcp

Configuration

Create a .env file (copy from example.env):

cp example.env .env

Environment Variables

Usage

Start the Server

python server.py

Or with uv:

uv run python server.py

Output:

=========================================
🚀 OpenCode MCP Server started
🎯 Target: http://127.0.0.1:4096
⏱️ Task timeout: 60 seconds
🔗 SSE URL:  http://127.0.0.1:14962/sse
=========================================

Available Tools

1. opencode_create_session

Create a new OpenCode session.

Input:

{
  "title": "My Task"  // optional
}

2. opencode_send_prompt

Send a prompt to an existing session and wait for result.

Input:

{
  "session_id": "ses_xxxx",
  "prompt": "Help me write a Python function"
}

3. opencode_check_session

Get session status and full conversation history.

Input:

{
  "session_id": "ses_xxxx"
}

4. opencode_list_sessions

List all sessions.

Input:

{
  "limit": 10  // optional, default 10
}

MCP Client Integration

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "opencode": {
      "command": "uv",
      "args": ["--directory", "/path/to/remote-opencode-mcp", "run", "python", "server.py"]
    }
  }
}

Other MCP Clients

Connect to the SSE endpoint:

http://localhost:14962/sse

Workflow Example

User → MCP Client → remote-opencode-mcp → OpenCode Server
                              ↓
                      1. Create session
                      2. Send prompt
                      3. Wait for result
                      4. Return to client

Development

Install Dev Dependencies

uv sync --extra dev

Linting

uv run ruff check .

Type Checking

uv run mypy .

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Related

• OpenCode - The AI coding agent

• MCP - Model Context Protocol