What your assistant can do once connected
Search your mail, read a message or a whole thread, look at attachments, mark things read, label, archive, create and disable aliases, and wait for a message that has not arrived yet. Eleven tools in total.
Echos speaks the connector protocol Claude already uses. There is no key to generate, copy, lose, or rotate.
Sign in with Google or with a code sent to your email. That is the only verification there is, and it happens before you see the dashboard, so there is nothing else to confirm before the next step.
In claude.ai open Settings, then Connectors, then Add custom connector. Name it Echos and give it the address https://api.echos.fyi/mcp. Leave the OAuth fields empty. Claude Desktop shows the same Connectors page.
Claude will show you what it is asking for: which of your addresses, and whether it may read, label, archive, or wait for mail. Approve it once.
"What is in my Echos inbox." If it answers, you are connected. Every call it makes from then on appears on your Agents page.
ChatGPT works too, with one extra step: you add Echos as an action inside a GPT of your own rather than approving a connector. Building a GPT with actions needs a paid ChatGPT plan. The step-by-step, with the values to paste, is on your Agents page once you are signed in.
Search your mail, read a message or a whole thread, look at attachments, mark things read, label, archive, create and disable aliases, and wait for a message that has not arrived yet. Eleven tools in total.
Send. Reply. Forward. Touch any other mailbox you own. Echos only knows about Echos.
There is a REST API, signed webhooks with retries, a rules engine, and an OAuth 2.0 server. On every plan, free included. The API is described by an OpenAPI document, the OAuth server publishes its metadata and accepts dynamic registration, and the MCP endpoint is https://api.echos.fyi/mcp. API keys, for scripts that are not OAuth clients, come from your Agents page.
Create your address, add the connector, approve it once.