opencode-cli-mcp

📖 Installation Guide — quick start, manual setup, and troubleshooting

MCP server wrapping opencode CLI's HTTP API (opencode serve) into 14 FastMCP tools. Also includes a FastAPI REST bridge, a Vite/React fleet-standard dashboard, and OpenCode custom tools that extend opencode itself.

Pattern: Plan with Claude, implement with opencode. Claude (expensive, high-judgment) orchestrates and supervises; opencode handles implementation grunt work on cheaper models (DeepSeek V4 Flash/Pro).

Quick Start

git clone https://github.com/sandraschi/opencode-cli-mcp
cd opencode-cli-mcp
just

This opens an interactive dashboard showing all available commands. Run just bootstrap to install dependencies, then just serve or just dev to start.

Manual Setup

If you don't have just installed:

Prerequisites

• opencode CLI: npm i -g opencode-ai

• Python 3.12+

• Node.js 18+

Run Everything

.\start.ps1 Starts: opencode serve (:4096) + FastAPI backend (:10951) + Vite frontend (:10950).

MCP Server Only

uv run -m opencode_cli_mcp.server Configure in Claude Desktop / Cursor / Windsurf (see Integration Guide).

14 MCP Tools

Key Workflows

See Usage Guide for full details.

Basic: Launch -> Poll -> Review

opencode_run_agent(prompt="refactor main.py", wait=false)
    -> { job_id: "abc" }
opencode_get_run_status("abc")     -> poll until completed
opencode_session_diff("session-xyz") -> review changes

Multi-Agent Sweep

Launch N agents across N repos in parallel, poll all, review diffs. Designed for fleet-wide operations.

Interactive Supervision

Start an agent, read its messages mid-task, send corrections, review final diff.

OpenCode Custom Tools

Copy .opencode/tools/*.ts into your opencode project to give opencode's LLM direct access to MCP fleet management, session inspection, and system diagnostics. 6 tools covering fleet, sessions, runs, system, providers, and tool discovery. See the OC Tools page in the webapp for full documentation and source.

Documentation

Ports

Security

This MCP server runs arbitrary shell commands (opencode run) from LLM prompts. Only install in environments where you trust your MCP client (Claude Desktop, Cursor) and the models it uses.

Fleet

• Registered in mcp-central-docs: ports 10950/10951

• fleet-registry.json and glama.json in repo root

• Webapp dashboard: http://localhost:10950 (run .\start.ps1)