MCP 服务器

OpenCode MCP Server

Integrates the OpenCode AI coding agent into MCP-compatible clients, allowing users to execute terminal-based coding tasks and manage sessions programmatically. It provides tools for running commands, listing AI models, and continuing existing coding sessions via the OpenCode CLI.

README

OpenCode MCP Server

An MCP (Model Context Protocol) server that provides seamless integration with OpenCode, the open-source AI coding agent for the terminal.

Features

Execute OpenCode Commands: Run any OpenCode CLI command programmatically
Session Management: Create, continue, and export coding sessions
Model Discovery: List available AI models from all configured providers
Async Execution: Non-blocking command execution with timeout handling
JSON Lines Parsing: Robust parsing of OpenCode's streaming output format

Tools Available

Tool	Description
`execute_opencode_command`	Execute any OpenCode CLI command with full flexibility
`opencode_run`	Run OpenCode with a simple prompt message
`opencode_continue_session`	Continue an existing OpenCode session
`opencode_list_models`	List available models, optionally filtered by provider
`opencode_export_session`	Export session data as JSON
`opencode_get_status`	Check OpenCode CLI availability and status

Installation

Prerequisites

Python 3.10+
OpenCode CLI installed and configured
MCP-compatible client (Claude Desktop, etc.)

Install Dependencies

pip install -r requirements.txt

Configure MCP Client

Add to your MCP client configuration (e.g., ~/.claude.json or Claude Desktop settings):

{
  "mcpServers": {
    "opencode": {
      "command": "python",
      "args": ["-m", "src.services.fast_mcp.opencode_server"],
      "cwd": "/path/to/opencode-mcp"
    }
  }
}

Usage

Basic Usage

Once configured, the MCP tools are available through your MCP client:

# Run a coding task
opencode_run(message="Create a Python function that calculates fibonacci numbers")

# List available models
opencode_list_models(provider="anthropic")

# Continue a previous session
opencode_continue_session(session_id="abc123", message="Now add unit tests")

# Check status
opencode_get_status()

Tool Parameters

`execute_opencode_command`

{
    "prompt": str,           # Required: The prompt/task for OpenCode
    "model": str,            # Optional: Model in provider/model format (e.g., "anthropic/claude-sonnet-4-20250514")
    "agent": str,            # Optional: Agent to use (e.g., "build", "plan")
    "session": str,          # Optional: Session ID to continue
    "continue_session": bool, # Optional: Whether to continue last session
    "timeout": int           # Optional: Timeout in seconds (default: 300, max: 600)
}

`opencode_run`

{
    "message": str,     # Required: Message/prompt to send
    "model": str,       # Optional: Model to use
    "agent": str,       # Optional: Agent to use
    "files": [str],     # Optional: Files to attach
    "timeout": int      # Optional: Timeout in seconds
}

`opencode_continue_session`

{
    "session_id": str,  # Required: Session ID to continue
    "message": str,     # Optional: Follow-up message
    "timeout": int      # Optional: Timeout in seconds
}

Configuration

Environment variables (prefix: OPENCODE_):

Variable	Default	Description
`OPENCODE_COMMAND`	`opencode`	Path to OpenCode CLI
`OPENCODE_DEFAULT_MODEL`	None	Default model to use
`OPENCODE_DEFAULT_AGENT`	None	Default agent to use
`OPENCODE_DEFAULT_TIMEOUT`	`300`	Default timeout (seconds)
`OPENCODE_MAX_TIMEOUT`	`600`	Maximum timeout (seconds)
`OPENCODE_SERVER_LOG_LEVEL`	`INFO`	Logging level

Architecture

src/services/fast_mcp/opencode_server/
├── __init__.py
├── __main__.py          # Entry point
├── server.py            # MCP server & tool definitions
├── opencode_executor.py # CLI execution wrapper
├── models.py            # Pydantic models
├── settings.py          # Configuration
└── handlers/
    ├── __init__.py
    ├── execution.py     # Run/continue operations
    ├── session.py       # Session management
    └── discovery.py     # Model/status discovery