The interesting engineering is not in this server — it is in the capture filters and the email-shaped vector index behind it, and what reaches the model is a single search call over the result. That is the right split for mail: the assistant never holds the archive, so the size of the mailbox stops being the limit. It only makes sense if your organisation already runs Mail2Cloud Advanced; the demo accounts exist for looking before you commit.
A Go client for mxHERO Mail2Cloud Advanced, a service that captures mail from one or more company accounts through filters, optimises it and stores it in an isolated tenant of a vector database built for email search. This server queries that tenant. Results come back with secure links to the original messages, which survive a user deleting their copy.
- `email_search` takes a query string and returns JSON search results across every captured account
- Secure links back to the original emails in the results
- Search over repositories far larger than a model's context window, because the retrieval happens in the vector store
- A native static binary — once compiled for your OS and CPU there are no runtime dependencies to install
An mxHERO Vector Search token; personal access tokens are issued at `https://lab4-mcp.mxhero.com/tokens`, and using the hosted service needs an account. Go 1.22 or higher to build — `go mod tidy` then a `make` target for your platform, which drops a binary in `bin`. Configure your client with the full path to the binary and pass the token as `-t`. Prebuilt binaries and signed installers are in the repository. This repo is the stdio build; a Streamable HTTP endpoint and a Python version exist separately.