The five tools cover the whole draft-to-publish loop, and deleting a draft before it goes out is there too, which matters when a model produces something you would rather not send. Because it is a single static binary the install is a copy rather than a dependency tree — the trade-off is that you build it yourself, as there is no package registry entry.
A Go rewrite of a WeChat Official Account MCP server, built on the mark3labs/mcp-go framework. It keeps the same tool names, arguments and responses as the earlier Python version, so it drops into an existing setup, and it ships as one binary or a container image.
- `WeChat_get_access_token` fetches the account access token the other calls need
- `WeChat_create_draft` creates an article draft
- `WeChat_publish_draft` publishes a draft to the account
- `WeChat_del_draft` deletes a draft that has not gone out yet
- `WeChat_del_material` removes a permanent media asset
- Structured logs go to stderr, including which tools registered and how each request was handled
Go source with a Makefile: `make build` produces `bin/wechat-oa-mcp`, or `make docker-run` builds and runs the container image `wechat-oa-mcp-go:latest`. Point your client at the binary path, or at `docker run -i --rm`. `LOG_LEVEL` controls verbosity — set it to debug when a call is failing.
One command — docker run -i --rm wechat-oa-mcp-go:latest