Give your agent
an outbound team

You get the same product either way, so pick on how you want to work. Use the MCP if you already live in Claude or ChatGPT, if what you want to work on is sitting in the conversation already, or if you would rather say a sentence than click through four screens. Use the app if you want to see everything laid out and click it yourself. Plenty of people use both.

Any MCP client. Claude, ChatGPT, Claude Code, Codex, Cursor and VS Code are the ones people use it in most. It is a remote server, so there is nothing to run locally and nothing to keep updated.

No. You sign in through your browser the first time your client connects and that is the whole setup. If your client would rather have a token, you can mint one under Settings, API: it carries only the permissions you tick and it is shown once.

No. The MCP is included with your plan and reading anything is free. Work that spends tokens, searching, research, finding a person, verifying an address, quotes the cost before it runs and refunds anything that comes back empty.

No. Every send is two calls that have to be separated by you answering: the first quotes how many real emails, to whom and when, the second sends. The server refuses to do both in one turn, so a model that tries anyway is stopped rather than obeyed.

Company descriptions, research answers and prospect replies all arrive tagged as untrusted with their source named. They are data about a stranger. Nothing inside one is ever executed as an instruction, however it is phrased.

Run npx @causo/mcp doctor. It reports what your token can do, whether a mailbox is genuinely ready to send, what your balance is, and whether anything is degraded at our end.