AI IDE Integration (MCP)

Jetic ships a stdio MCP server (jetic mcp) that gives the AI assistant inside your editor typed tools to inspect, author, validate, and live-test your API model and workflows.

It uses your editor's own model — there is no OpenRouter/OpenAI key, no jetic config ai, and no jetic scan required to begin. Because the AI reads your code directly, this path also works for any backend stack (Express, FastAPI, Go, Java, NestJS, …).

jetic mcp

How project resolution worksCopy Anchor Link

Every tool accepts an optional projectPath. Resolution order:

If no model.json exists, read tools return modelExists: false with a hint instead of pretending an empty model is real. Run jetic init + jetic scan first in that case.

Editor setupCopy Anchor Link

All entries are local stdio servers. After adding the config, restart the editor (or refresh its MCP panel) and look for the jetic_ tools.

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "jetic": {
      "type": "local",
      "command": ["node", "C:/Users/you/Jetic/packages/mcp-server/dist/index.js"],
      "environment": { "JETIC_PROJECT_PATH": "C:/path/to/my-api" },
      "enabled": true
    }
  }
}

"command": ["jetic", "mcp", "--project", "C:/path/to/my-api"]

opencode mcp list

{
  "mcp": { "jetic": { "type": "local", "command": ["jetic", "mcp"], "enabled": true } },
  "tools": { "jetic_*": false },
  "agent": { "build": { "tools": { "jetic_*": true } } }
}

Onboarding promptsCopy Anchor Link

Paste these into your editor's agent chat, in order.

1. Meet Jetic (first message)

   I've installed the Jetic MCP tools. Summarize what you can do with them, then read my project's model summary and tell me what's in it.

2. Re-analyse the whole project

   Help me re-analyse the whole project: find ALL API endpoints — routes, controllers, middleware, auth guards, request/response shapes — and add every missing endpoint to .jetic/model.json using the jetic tools. Include security, middleware chains, parameters, and response schemas. When done, run jetic_verify_model and fix all errors.

3. Zoom into one area

   Analyse ONLY src/modules/payments: add/update all of its endpoints with full detail (middleware, auth, validation constraints), then verify.

4. Create a workflow

   Create a workflow called "Shopper checkout": register a user with a faker email and capture the credentials, log in and capture the access token, then create an order with the Bearer token expecting 201. Validate it BEFORE saving, save it, and simulate it against http://localhost:3000.

5. Live-test loop

   Test POST /api/auth/login against my local server at http://localhost:4000 with a realistic payload and tell me whether the response matches the model. Then verify the whole model and fix any issues.

Behind the scenes the agent follows the structured order: jetic_init → jetic_scan (or jetic_add_endpoint) → jetic_verify_model → jetic_validate_workflow → jetic_create_workflow → jetic_simulate_workflow. If a step fails, paste the error back — validation messages name the exact fix.

MCP toolsCopy Anchor Link

The server advertises machine-readable annotations (readOnlyHint, destructiveHint, openWorldHint) and startup instructions that nudge agents through initialize → model → simulate. Tool calls execute one at a time in arrival order, so agents may fire parallel calls safely.

Session & change trackingCopy Anchor Link

Alongside the core tools:

• jetic_get_session_phase — returns the current phase (init → scan → verify → simulate) and the next action, so an agent can self-orient at the start of a session.
• jetic_get_changes — reads .jetic/changes.json to re-index only edited files instead of the whole codebase.
• jetic_clear_changes — clears the change log after the model is updated.

Authoring loopCopy Anchor Link

TroubleshootingCopy Anchor Link