# MCP for agents > Give an agent narrowly scoped tools for listing, reading, sending, and replying to mailbox conversations. ## Endpoint and authentication Startup Mail exposes a stateless Streamable HTTP MCP endpoint at: ```text https://api.startupmail.dev/mcp ``` Use the same Bearer API key as the REST API. Available tools are determined when the connection is created from that key’s scopes. ## Configure a client The exact configuration format depends on the MCP client. A typical HTTP entry needs the URL and authorization header: ```json { "startupmail": { "type": "http", "url": "https://api.startupmail.dev/mcp", "headers": { "Authorization": "Bearer ${STARTUPMAIL_API_KEY}" } } } ``` Do not paste a production key into a configuration file that will be committed or shared. Prefer environment-variable interpolation supported by your client. ## Available tools | Tool | Required scope | Purpose | | ------------------ | ---------------- | --------------------------------------------------- | | `list_mailboxes` | `mailboxes:read` | See addresses available to the key | | `list_threads` | `mail:read` | List recent threads for a mailbox and label | | `get_thread` | `mail:read` | Read a thread with messages and attachment metadata | | `send_email` | `mail:send` | Start a conversation from an accessible mailbox | | `reply_to_message` | `mail:send` | Continue a thread from a specific message | ## Permission an agent safely Create a dedicated key for one agent or workload. Start with read-only scopes, add `mail:send` only when the workflow needs it, and keep human approval around ambiguous or high-impact sends. > **Scope is necessary, not sufficient** > > Your agent instructions and application should still restrict recipients, sending purpose, and > when a human must approve. Startup Mail’s acceptable use policy applies to agent actions exactly > as it does to human sends.