WeApp Pilot MCP

WeApp Pilot MCP is a staged implementation of a local MCP server for LLM-driven smoke testing of WeChat Mini Programs.

The current scope is intentionally narrow:

Taro-first project detection and build orchestration
local-machine WeChat DevTools sessions
miniprogram-automator backed page actions
MCP tools for session, interaction, assertion, and artifact flows

Packages

@weapp-pilot/shared: contracts, error model, and tool schemas
@weapp-pilot/weapp-core: local core logic, build orchestration, runtime actions, assertions, and run artifacts
@weapp-pilot/mcp-server: stdio MCP server and tool registry

Tool Surface

The current MCP server exposes four groups of tools:

Session: project_detect, project_routes, route_search, build_start, build_status, ide_open, ide_status, automator_connect, runtime_disconnect
Interaction: navigate, current_page, page_stack, tap, input, page_data, page_snapshot, call_wx, screenshot
Assertion: assert_route, assert_text, assert_visible, assert_page_data
Artifact: console_logs, network_logs, run_summary
Scenario: smoke_run

project_routes reads the compiled app.json route catalog and now enriches each route with origin and discovered page titles from compiled page JSON files. route_search ranks candidate routes by route segments and page titles so Agents can resolve targets before calling smoke_run. page_snapshot compresses the current page into route, page stack, visible texts, data paths, and auth signals. console_logs persists runtime console and exception events, and network_logs now records patched wx.request / uploadFile / downloadFile activity after the runtime session connects.

ide_open can now infer miniProgramRoot from project detection when callers only provide projectRoot. Route assertions and smoke_run also normalize leading slashes, so /pages/foo/index and pages/foo/index are treated as the same runtime route. Runtime log artifacts are persisted in a default-safe form: network_logs and console text redact common secrets such as tokens, cookies, passwords, verification codes, and phone values while preserving host/path/status metadata for debugging.

Quick Start

pnpm install
pnpm build
pnpm test
pnpm smoke:fixture

pnpm smoke:fixture runs the end-to-end fake smoke validation around packages/weapp-core/test/fixtures/taro-smoke-demo, including route navigation, interaction, assertions, screenshot capture, and run summary persistence.

MCP Server

Start the stdio server with:

pnpm mcp:start

Typical MCP client wiring looks like:

{
  "mcpServers": {
    "weapp-pilot": {
      "command": "pnpm",
      "args": ["mcp:start"],
      "cwd": "C:/Users/admin/Documents/GITHUB/WeApp-Pilot"
    }
  }
}

Local Validation

Real-machine validation still requires a prepared WeChat DevTools environment:

a logged-in WeChat DevTools installation
the DevTools service port enabled
a resolvable DevTools CLI path
a Taro project that builds to a Mini Program output directory

The step-by-step setup and validation flow is documented in docs/local-setup.md.

MCP Servers