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.
Links
README
From the repo.
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: "<available-model>"
})
// 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
Project-scoped tools accept directory: an absolute path on the OpenCode server. POSIX, Windows drive, and UNC paths are preserved across client operating systems. Relative paths are rejected; OpenCode checks existence and access.
opencode_project_init({path: "/absolute/local/project"}) creates or opens a directory on the MCP host. For remote OpenCode servers, create the project on that server instead. Authentication tools are global. Resources offer both static reads of the default project and explicit project/session templates.
Independent sessions do not isolate filesystem changes. Use separate project directories or Git worktrees when running overlapping coding tasks in parallel.
Configuration
All settings are optional; an OpenCode server must already be running by default.
| Variable | Purpose |
|---|---|
OPENCODE_BASE_URL | Server endpoint; defaults to http://127.0.0.1:4096 |
OPENCODE_SERVER_USERNAME, OPENCODE_SERVER_PASSWORD | Optional server HTTP authentication |
OPENCODE_AUTO_SERVE | Set to true to opt in to launching a local server |
OPENCODE_DEFAULT_PROVIDER, OPENCODE_DEFAULT_MODEL | Default prompt provider/model pair |
OPENCODE_TOOL_PROFILE | full (default) or a smaller essential tool set |
OPENCODE_TASK_STORE | Override the local directory for persisted job records |
See configuration for storage, permissions, and client setup. Auto-start only supports local loopback HTTP endpoints; custom OpenCode CLI flags require a manually started server.
Development and Verification
npm ci
npm test
npm run test:coverage
Tests use local fixtures and do not require a model subscription. For controlled live checks against a local OpenCode server:
npm run build
node scripts/mcp-smoke-test.mjs
Live smoke checks use a disposable project and owned session. Inference is opt-in with an explicitly selected provider and model. See live verification and releases for scope, skipped capabilities, and publishing checks.
Documentation
Collected info
- ★ 138 stars
- ⎇ 25 forks
- Language: TypeScript
- Source updated: 9/22/2026
Config for your environment
Replace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.
Tool
OS
Config file: ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "{MCP_ENDPOINT_URL}"
}
}
}Paste into mcpServers in the config file. Restart Cursor after saving.
If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.