The context and the files travel with the question, so the consulted model starts from the same material rather than from a summary — and the file-writing path covers the models you can only reach through a browser.
A consultation channel between models. It packages the context you nominate — including whole files — sends it to another model through OpenRouter, and returns the reasoning for the model you are talking to to act on.
- A problem sent to an external model for deep reasoning, with the related context and a list of files attached
- A manual path for the same thing: a master prompt written to a file you choose, for pasting into a model you cannot reach by API
- A choice of model through OpenRouter, so the consultant can be picked per problem rather than fixed
- The response returned as something the calling model can use, not as a transcript to read
Uvx, which fetches and runs the published package without cloning anything, and an API key for the provider you are routing through, passed on the launch command. Calls to the external model are billed by that provider.
One command plus a key — uvx phone-a-friend-mcp-server, then supply credentials
