People with a personal agent
Your OpenClaw, Hermes or Claude Code gets its own email. It receives sign-up codes and reads the inbox, and you stop lending it your Gmail.
An inbox for your AI agent: Claude Code, OpenClaw or Hermes.
Replyz is the email platform for AI agents. Your agent gets its own address and receives sign-up codes and confirmation links. Every email is scored for prompt injection, and the account will belong to a human you choose (claiming is coming soon). EU data storage.
Works with
- Claude Code
- OpenClaw
- Hermes Agent
- Codex
- Cursor
- Any MCP client
Who is Replyz for?
Developers building agents
An inbox per agent or task (coming soon: 3 live inboxes per claimed account), through the API, the CLI and the MCP server; the SDKs are in progress. wait and extract are single calls, not a parser you maintain.
Teams that need the EU
Mail is stored at rest in the EU and processed on Cloudflare's global network. A data processing agreement (DPA) is planned.
Give your agent email, whichever agent it is
Via Claude Code
Install the skill and ask: "get yourself an inbox and give me the confirmation link for this newsletter". The CLI keeps the API key out of the model's context.
Via OpenClaw
Add the Replyz skill, then message your agent from Telegram or WhatsApp. It reads the email side of every errand.
Via Hermes Agent
Same skill, same CLI. Hermes gets an address of its own instead of logging in to yours.
Via Claude Code (and claude.ai in the request-header beta)
In Claude Code, add Replyz as a remote MCP server with an API key header. In claude.ai, Claude Desktop and Cowork, request headers are a beta for some organisations, so most people need OAuth, which is planned. ChatGPT needs OAuth too.
Via Cursor and any MCP client
One endpoint, standard Streamable HTTP. Point any MCP client at it and the tools show up.
Via the API and SDKs
REST with long-polling wait. JavaScript and Python SDKs are in progress.
What your agent can do with its own inbox
Get past the code step
Fill in the form, wait for the confirmation, finish the sign-up. No human copying codes. Use it only where a service's terms allow automated sign-ups.
Read codes and links
One call returns the verification code or link from the right email, even if it arrived first.
Read incoming mail
Suppliers, bookings, support threads: the agent reads them and tells you what matters.
Receive-only
Your agent can't send email.
Stay in the EU
Mail is stored at rest in the EU and processed on Cloudflare's global network. A DPA is planned.
Use open-source clients
The CLI and skill are MIT-licensed on GitHub; the SDKs are in development. The email service itself is hosted by Replyz.
Safe by default
- Every email is untrusted. Each message is marked as untrusted content and carries an injection-risk score with the signals behind it, so your agent can tell when an email is trying to give it orders.
- Most hidden text is removed. Most hidden HTML and invisible characters are removed (heuristics with known gaps), so treat every email as untrusted anyway.
- Nothing is fetched. Replyz never opens links or loads images, and attachments are download-only, with risky ones flagged.
- Every account is meant to have a human owner. The agent signs up; you claim the account (claiming is coming soon).
- Receive-only. Your agent can't send email.
These are heuristics that lower the risk; they don't make email safe. Agents should treat email as data, never as instructions.
How it works
- 1
Your agent signs up
It runs the CLI or the skill and gets an API key and a first inbox. No form for you to fill.
- 2
You claim the account (coming soon)
Once claiming opens, you claim the account by sending one email from your own address. That ties it to you and lifts the sandbox limits.
- 3
It works on its own
It waits for email, reads codes and links. It can't send email: Replyz is receive-only.
Paste this into your agent
Get yourself an email inbox: run `npx -y @replyz/cli signup --name <you>`. Save the api_key and tell me the address. Docs: https://replyz.io/llms.txt
Full steps in the quickstart.
Or connect the MCP server
claude mcp add --transport http replyz https://mcp.replyz.io/mcp \ --header "Authorization: Bearer $REPLYZ_API_KEY"
Frequently asked questions
Can my AI agent have its own email address?
Yes. Replyz gives each agent its own inbox and address. Agents get an inbox through the CLI, skill or MCP server, so they never need your personal Gmail.
How does Replyz protect my agent from prompt injection in emails?
Every message is cleaned and scored: most hidden HTML and invisible characters are removed (heuristics with known gaps), each message gets an injection-risk score with the signals behind it, risky attachments are flagged, and links and images are never fetched. These are heuristics that lower the risk; agents should still treat email as data, never as instructions.
How does my agent read a verification code?
It waits for the matching email in one call and asks Replyz for the code or link, even if the email arrived before the call. Both are available through the API, the CLI and the MCP server. Use it only where a service's terms allow automated sign-ups.
Who owns the account, and can the agent send email?
Every account is meant to have a human owner: the agent signs up, then you claim it by sending one email from your own address (claiming is coming soon). Replyz is receive-only, so the agent can't send email. Sending is not available.
Which agents will work with Replyz?
Claude Code, OpenClaw and Hermes through the skill or CLI, and Codex, Cursor or any MCP client that can send an API key header through the MCP server. Claude on the web, Desktop and Cowork needs request headers, a beta for some organisations. ChatGPT needs OAuth, which is planned.
Why not just give my agent my Gmail?
A Gmail connection gives the agent your whole inbox and lets it send as you. For sign-ups, codes and errands, the agent only needs an address of its own.
How is Replyz different from AgentMail?
Both give agents inboxes. Replyz adds a call that waits for the next email and built-in code extraction, scores every message for prompt injection, will tie every account to a human (claiming is coming soon), stores mail at rest in the EU. AgentMail has been live for longer; Replyz is brand new.
Does it work with LangChain, n8n or the Vercel AI SDK?
Replyz is an HTTP API, so any framework that can make HTTP requests can use it, and agents can also use the CLI. There are no native framework integrations yet.
What about attachments?
Attachments on incoming mail are stored and served as downloads only, and risky types are flagged. Replyz cannot send email or attachments.
Where is my data stored?
Mail and account data are stored at rest in the EU; requests are processed on Cloudflare's global network, which may be outside the EU. Replyz is operated by a Portuguese company. A data processing agreement is planned.
Is Replyz open source?
The CLI and agent skill are public and MIT-licensed on GitHub; the SDKs are in development. The email service itself is hosted by Replyz.
Is Replyz a temporary or disposable email service?
No. Inboxes are meant to belong to an account a human claims (coming soon); sandbox accounts expire after 7 days. Replyz does not send email. Replyz is built for agents that do real work, not for throwaway sign-ups.
Give your agent an inbox.
Your agent signs itself up. Start with the quickstart.