What it does
weapp-agent-mcp is an MCP server built on miniprogram-automator. It lets agents such as Claude Code, Cursor and Codex control WeChat DevTools directly within a conversation, without writing intermediate scripts. Available tools cover connection checks, navigation, screenshots, page snapshots, element taps, calling wx APIs, evaluating code in the AppService, polling conditions, reading console logs, and running short test scenarios that produce markdown reports. Health-check and recovery tools help restore a dropped connection during long sessions.
It suits agent-led debugging and page inspection, serial screenshots, lightweight regression scenarios, and checks that rely on stable selectors. It is not designed for concurrent screenshots, stress testing, or very long end-to-end chains in CI. For conventional scripted end-to-end tests, the miniprogram-automator SDK used directly is lighter.
Running it requires WeChat DevTools with command-line access, Node.js 18 or later with npm, a Mini Program project that can be opened in DevTools, and the automation and service port enabled in the DevTools security settings. It can be added to an MCP client through npx or built from source. It either connects to an already open IDE through an automation WebSocket endpoint, or launches the IDE through the CLI for a given project path. Environment variables control the connection mode, ports and timeouts.
Limitations and requirements
Requires WeChat DevTools with CLI access, Node.js 18+, npm, a Mini Program project, and automation and service ports enabled in DevTools. Screenshots run on a serial channel and are not suited to concurrent load. Complex actions can time out, so long scenarios should be split into shorter ones. Run the health check first, and run recovery only when it reports that recovery is needed. Reliability depends on the app providing clear qa-* selectors or other stable anchors. Some clients cannot reach the DevTools CLI outside the project directory, so only the WebSocket connection is supported in those cases. Claude Code may need tool permissions pre-approved, otherwise the connection can drop during prompts.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
io.github.Chaixueyuan/weapp-agent-mcp
Agent-first MCP server for WeChat Mini Program debugging & automation via WeChat DevTools.
v0.4.8As shown on registry.modelcontextprotocol.io when we last read it