The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Opencode MCP
MCP server for OpenCode AI — 70 tools, 10 resources, 5 prompts. Use npx opencode-mcp with Claude Desktop, Claude Code, Cursor, Windsurf, or any MCP client.
Install
claude mcp add opencode -- npx -y opencode-mcpInstall command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →opencode-mcp
Delegate coding work to OpenCode from your MCP client.
opencode-mcp connects Claude, Cursor, VS Code, and other MCP clients to OpenCode's headless API. Ask questions, implement features, monitor background work, respond to questions and permissions, and review changes across projects.
Version 3.0.0 requires Node.js 22 or newer. Upgrading from 2.x? See the migration notes.
Quick Start
Install OpenCode and start its server from your project:
opencode serve --hostname 127.0.0.1 --port 4096
If you use the TUI, start it with opencode --port 4096 and share that server. Set OPENCODE_BASE_URL for another endpoint.
For Claude Code:
claude mcp add opencode -- npx -y opencode-mcp
For clients using an mcpServers configuration:
{
"mcpServers": {
"opencode": {
"command": "npx",
"args": ["-y", "opencode-mcp"]
}
}
}
Restart the client and call opencode_setup. Choose a provider from its configured providers, then use opencode_provider_models to select a model. Set OPENCODE_DEFAULT_PROVIDER and OPENCODE_DEFAULT_MODEL together or pass the selected IDs in each prompt call.
Client-specific configuration includes VS Code, Windsurf, Continue, Zed, and Amazon Q. To test unreleased changes, build from source and configure your client to run node with the absolute path to dist/index.js.
Choose a Workflow
| Need | Tools |
|---|---|
| Setup and orientation | opencode_setup, opencode_context, opencode_provider_models |
| Quick question or follow-up | opencode_ask, opencode_reply |
| Start work and wait | opencode_run |
| Work in the background | opencode_fire, then opencode_check or opencode_wait |
| Recover or control a recorded job | opencode_job_list, opencode_job_get, opencode_job_cancel |
| Resolve required input | opencode_job_input, permission and question tools |
| Review the result | opencode_review_changes, opencode_conversation |
opencode_fire({
directory: "/absolute/path/to/project",
prompt: "Add input validation to POST /api/users and run the relevant tests",
providerID: "<configured-provider>",
modelID: ""
})
// Save the returned job and session IDs; use them to monitor or resume observation.
Async results distinguish accepted, running, input_required, completed, failed, cancelled, and unknown. An observation timeout returns current progress; it does not mean the task failed or was cancelled. Follow the returned state and IDs instead of assuming an absent busy status means success.
Modern clients can use the MCP Tasks extension for opencode_run. Clients without that extension use ordinary tools, including opencode_fire and opencode_check. Task status is retrieved by polling; this package does not promise to wake an idle assistant with completion notifications.
Tools retain readable text and provide structured results for clients that consume them. Prompt tools accept optional model variants and OpenCode structured-output formats. See the generated tools reference and examples.
Multi-Project Use
Similar plugs
Reverse engineer anything with agents, from app behavior down to native binaries.
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype…
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io