Dev / Test Plugins

dryrunPlugin

Skips handler execution and returns parameter validation + expected result info. For testing, CI, and schema verification.

import { dryrunPlugin } from '@airmcp-dev/core';

// Global mode — all calls are dry-run
use: [dryrunPlugin({ enabled: true })]

// Environment variable control
use: [dryrunPlugin({ enabled: process.env.DRY_RUN === 'true' })]

// Per-call mode — only calls with _dryrun parameter
use: [dryrunPlugin({ perCall: true })]

Option	Type	Default	Description
enabled	boolean	false	Global dry-run activation
perCall	boolean	true	Only dry-run calls with _dryrun: true param
mockResponse	(toolName, params) => any	—	Custom dry-run response generator

Default dry-run response

{
  "dryrun": true,
  "tool": "search",
  "params": { "query": "hello" },
  "description": "Search documents",
  "schema": [
    { "name": "query", "type": "string", "provided": true, "value": "hello" },
    { "name": "limit", "type": "number", "provided": false, "value": null }
  ],
  "message": "[Dryrun] \"search\" would be called with: {\"query\":\"hello\"}"
}

Custom response

use: [dryrunPlugin({
  enabled: true,
  mockResponse: (toolName, params) => {
    if (toolName === 'search') return { results: [], total: 0 };
    return `[Mock] ${toolName} would run`;
  },
})]

perCall mode caveat

perCall mode only works with server.callTool() direct calls:

// ✅ Works — callTool passes through middleware chain as-is
server.callTool('search', { query: 'hello', _dryrun: true });

// ❌ Doesn't work — MCP SDK strips params not in schema
// Claude Desktop → server calls won't have _dryrun param

For MCP client calls, use enabled: true global mode or environment variables.