Skip to content

examples.md

Latest commit

 

History

History
194 lines (145 loc) · 4.91 KB

examples.md

File metadata and controls

194 lines (145 loc) · 4.91 KB

Usage Examples

Real-world examples of using opencode-mcp tools from any MCP client.

Quick Question

Ask OpenCode something and get an answer in one call:

opencode_ask({
  "prompt": "Explain the authentication flow in this project",
  "providerID": "anthropic",
  "modelID": "claude-opus-4-6"
})

Build a Feature

Have OpenCode implement something and wait for it to finish:

opencode_run({
  "prompt": "Add input validation to POST /api/users. Validate email format, non-empty name, and positive integer age. Return 400 with descriptive errors.",
  "providerID": "anthropic",
  "modelID": "claude-opus-4-6",
  "maxDurationSeconds": 300
})

Then review what it did:

opencode_review_changes({ "sessionId": "<session-id>" })

Multi-Turn Conversation

Start a session and iterate:

1. opencode_ask:

{ "prompt": "What testing framework does this project use?", "title": "Testing exploration" }

2. opencode_reply:

{ "sessionId": "<session-id>", "prompt": "Show me an example of writing a new test" }

3. opencode_reply:

{ "sessionId": "<session-id>", "prompt": "Now add a test for UserService.create" }

Background Tasks (Fire-and-Forget)

Dispatch a long-running task and keep working on something else:

1. Fire the task — opencode_fire:

{
  "prompt": "Refactor the entire authentication module to use JWT tokens",
  "providerID": "anthropic",
  "modelID": "claude-opus-4-6",
  "title": "JWT refactor"
}

Returns immediately with a session ID.

2. Check progress — opencode_check:

{ "sessionId": "<session-id>" }

Returns: status (running/idle), todo progress (e.g. "3/7 done, current: Add JWT middleware"), files changed count.

3. Get full results when done — opencode_conversation:

{ "sessionId": "<session-id>" }

Build a Full App

Chain multiple tools to build an entire project:

// 1. Set up the project
opencode_run({
  "prompt": "Create a new Express.js API with TypeScript, SQLite, and Vitest. Include health check, CRUD for users, and error handling middleware.",
  "title": "Build API",
  "maxDurationSeconds": 600
})

// 2. Add features in parallel
opencode_fire({ "prompt": "Add JWT authentication with login/register endpoints", "title": "Add auth" })
opencode_fire({ "prompt": "Add rate limiting middleware and request logging", "title": "Add middleware" })

// 3. Check both are done
opencode_check({ "sessionId": "<auth-session-id>" })
opencode_check({ "sessionId": "<middleware-session-id>" })

// 4. Review all changes
opencode_review_changes({ "sessionId": "<auth-session-id>" })
opencode_review_changes({ "sessionId": "<middleware-session-id>" })

Get Project Context

Understand a project you've never seen:

opencode_context({})

Returns: project info, VCS details (branch, status), configuration, and available agents.

Code Review

Review changes from a coding session:

1. Find the session — opencode_sessions_overview:

{}

2. Review the diff — opencode_review_changes:

{ "sessionId": "<session-id>" }

3. Read the conversation — opencode_conversation:

{ "sessionId": "<session-id>" }

Search the Codebase

// Find TODOs and FIXMEs
opencode_find_text({ "pattern": "TODO|FIXME|HACK" })

// Find config files
opencode_find_file({ "query": "config" })

// Find a function or class
opencode_find_symbol({ "query": "handleAuth" })

Multi-Project Workflow

Work on multiple projects from one client:

// Mobile app
opencode_ask({
  "directory": "/home/user/projects/mobile-app",
  "prompt": "Set up React Navigation with a tab navigator"
})

// Web app (same server, different project)
opencode_ask({
  "directory": "/home/user/projects/web-app",
  "prompt": "Add authentication to the Next.js app"
})

Provider Management

// Check what's available
opencode_provider_list({})

// Set an API key (one-time, global)
opencode_auth_set({
  "providerId": "anthropic",
  "type": "api",
  "key": "sk-ant-..."
})

// Test that it works
opencode_provider_test({ "providerId": "anthropic" })

Using Prompts

MCP prompts are guided workflow templates your client offers as selectable actions.

Prompt Arguments What it does
opencode-code-review sessionId Reviews diffs for correctness, style, performance, security
opencode-debug issue, context? Step-by-step debugging: finds files, reads code, identifies root cause
opencode-project-setup (none) Reads README, configs, entry points, summarizes the project
opencode-implement description, requirements? Sends to OpenCode's build agent, reviews changes, reports results
opencode-best-practices (none) Guides on setup, tool selection, monitoring, and common pitfalls
opencode-session-summary sessionId Summarizes discussion, actions, files modified, remaining work