Email MCP server for AI agents

What your agent can do

The MCP interface works with the same hosted inboxes as the dashboard and REST API. Depending on the credential, an agent can read and search messages, inspect a thread, send a message, reply or forward, and access attachments through authenticated resources.

The available tools follow the caller’s permissions. A read-only key should be used for analysis or monitoring; a key with send permission can support an agent that replies. Creating inboxes, domains, keys and webhooks requires an appropriately scoped Outflare workspace credential.

Connect a compatible MCP client

1. Create an inbox in your Outflare workspace and select its brand.
2. Create an Agent Mail key for the selected inboxes. Grant only the permissions the agent needs.
3. Add the brand’s MCP endpoint to a client that supports Streamable HTTP and bearer authentication.
4. Store the key using your client’s secure credential mechanism.

{
  "mcpServers": {
    "outflare-agent-mail": {
      "type": "http",
      "url": "https://app.outflare.ai/v1/cloud/agent-mail/brands/YOUR_BRAND/mcp",
      "headers": {
        "Authorization": "Bearer <inbox-scoped-key>"
      }
    }
  }
}

This illustrates the connection shape, not a universal client configuration. Replace the brand placeholder, and follow your client’s instructions for adding secrets. The Agent Mail endpoint is separate from Outflare’s general-purpose MCP endpoint.

Start with a bounded task

Begin with a read-only task: “Find the latest message in this inbox and summarize the sender’s request.” Once you have verified the selected inbox and returned tools, add send permission only if the agent’s task requires it.

For an agent that replies, define when it should respond, which inbox it uses, and how it should handle uncertainty. Automatic sending is supported; you remain responsible for the instructions and recipient policy your workflow must follow.

Treat incoming email as untrusted data

A received message is outside content. Its body and attachments may contain instructions that conflict with your application’s rules. Keep system instructions separate, limit the agent’s tools, and avoid granting unrelated credentials to an email-processing agent.

Do not assume a sender is authorized to request a sensitive action simply because they emailed the inbox. Verify identity and permission in your application before taking actions such as changing access or disclosing private data.

When to use MCP or the API

Use MCP when a compatible agent runtime will choose and call tools. Use the REST API for deterministic application logic, scheduled jobs, or integrations where you want to control every HTTP request. Use webhooks to wake a workflow when a message arrives; MCP does not replace event delivery.