Replyz
आपके AI एजेंट का अपना इनबॉक्स: साइन-अप कोड और लिंक। यह केवल अपने मानव को ईमेल कर सकता है। कुंजी आवश्यक है।
होस्ट किया गया MCP सर्वर
npx add-mcp 'https://mcp.replyz.io/mcp'Claude Code, Codex, Cursor और अन्य में इंस्टॉल होता है
दस्तावेज़
An email MCP server that gives your agent its own inbox.
Most email MCP servers connect your agent to your Gmail or Outlook. Replyz gives the agent an inbox of its own, so it can receive sign-up codes and confirmation links, without access to your personal mail. Use it only where a service's terms allow automated sign-ups.
How do you connect it?
Remote MCP endpoint, Streamable HTTP, authenticated with your API key (or, in claude.ai, Claude Desktop and Cowork, by signing in with OAuth: see Replyz for Claude):
https://mcp.replyz.io/mcp
Authorization: Bearer $REPLYZ_API_KEY
Claude Code
claude mcp add --transport http replyz https://mcp.replyz.io/mcp \
--header "Authorization: Bearer $REPLYZ_API_KEY"
Or let the CLI set it up
npx -y @replyz/cli mcp install
Which tools does it have?
create_inbox, list_inboxes
Another address once the account allows it (sandbox: 1 live inbox, up to 3 once the account is claimed).
wait_for_email
Waits up to 45 s for a matching email, by sender, subject or time, then returns it.
get_verification
Returns the code or link from the latest matching email, even if it arrived before the call.
list_messages, read_message
Read the inbox. Email content is always marked as untrusted.
email_my_human
The agent queues a short email to you, the human who claimed the account, to report back. Up to 10 a day; delivery is not guaranteed.
reply_to_email
One plain-text reply to the sender of a message the inbox received; there is no recipient input. Listed only with an API key, for claimed accounts on the 21 Nov 2026 terms or later; OAuth connectors do not list it. Up to 50 a day per account, 20 per inbox, 10 per conversation and 50 per owner; queued, not guaranteed to arrive.
get_account
Limits, usage and whether the account is claimed.
Code and link extraction: 91.7 % on a 48-mail blind test sealed before the fixes (measured 18 Nov 2026). Where it still misses: see the write-up.
What does a session look like?
you Subscribe to the newsletter and tell me when it's confirmed.
agent list_inboxes → newsletter-quiet-finch-7c1e2k@youreplyz.com
agent get_verification(from: "*@news.example")
✓ link: https://news.example/confirm?t=…
Why not use a Gmail MCP server?
- Scope. A Gmail MCP server hands the agent your whole inbox. With Replyz it only sees mail sent to its own address.
- No Google OAuth app. Nothing to approve on your Google account, and nothing to revoke later.
- Built for agents. Waiting for an email and pulling out a code are single tool calls.
- EU data storage. Mail is stored at rest in the European Union. Requests are processed on Cloudflare's global network, which may be outside the EU.
Related guides
- Add an MCP server to Claude Code (claude mcp add)
- Email MCP servers compared
- Keep an API key out of the agent's context
- How an agent gets a verification code from email in code
Frequently asked questions
What is an email MCP server?
An MCP server that gives an AI agent email tools. Most connect the agent to your Gmail or Outlook. Replyz gives the agent an inbox of its own instead.
How do I connect the Replyz MCP server?
Point your MCP client at https://mcp.replyz.io/mcp over Streamable HTTP and send your API key as a Bearer token, or, in claude.ai, Claude Desktop and Cowork, add it as a custom connector and sign in with OAuth. The Replyz CLI can also do it for you with replyz mcp install.
Which tools does the Replyz MCP server have?
create_inbox, list_inboxes, list_messages, read_message, wait_for_email, get_verification and get_account. email_my_human queues one short email to the human who claimed the account, up to 10 a day (delivery is not guaranteed). reply_to_email replies to the sender of a received message; it is listed only with an API key, for claimed accounts on the 21 Nov 2026 terms or later, and OAuth connectors do not list it.
How long does wait_for_email wait?
Up to 45 seconds per call. If nothing arrives it returns a timeout and the agent simply calls it again.
Is email content safe to give to the model?
Every tool marks email content as untrusted, so the agent treats instructions inside an email as data, not as commands.
Should I use the MCP server or the CLI?
Use the CLI or the skill for agents that run on your machine (Claude Code, OpenClaw, Hermes): it is faster and keeps the API key out of the model's context. Use the MCP server for remote agents such as Cursor. Claude on the web, Desktop and Cowork use a custom connector with OAuth sign-in. ChatGPT: not tested yet. Replyz now has OAuth sign-in, which ChatGPT connectors need, but we have tested it only with Claude.
My agent runs in a sandbox that only allows some domains. Will Replyz work?
Only after the owner allows api.replyz.io (and mcp.replyz.io). Agents in sandboxes that only allow listed domains, such as some cloud agents and code-execution environments, cannot reach Replyz until their human adds those domains to the allowlist.