Integrations · OpenClaw
Give your OpenClaw agent its own inbox
Install the Pidgeon channel and sign in, and email to your agent’s address starts an OpenClaw conversation and is answered in the thread.
01 · Set it up
5 steps, about two minutes.
01 · Install the channel
Pure TypeScript for OpenClaw 2026.10 or later, run on macOS and installed on Linux x64 and arm64. OpenClaw asks you to confirm it and accept its capabilities; a script passes --force --accept-capabilities.
openclaw plugins install @pidgeon-ai/openclaw02 · Turn it on and sign in
The login opens Pidgeon in your browser and waits ten minutes for you. You choose what it can do — read, read and send, or full access — and which address it reaches, or every address if you own the account. That becomes an API key under Developers → API keys → Connected apps, and Disconnect there ends it. The channel only needs read and send.
openclaw config set channels.pidgeon '{}' --strict-json openclaw channels login --channel pidgeon03 · Check it
Says running and works when it is reading mail. Send the address an email: the agent answers in the thread, and a reply to that continues the same conversation.
openclaw channels status --channel pidgeon --probe04 · Or connect with a key
For an OpenClaw on a server with nobody there to sign in. The channel reads PIDGEON_API_KEY from the Gateway’s environment, never from the config file.
export PIDGEON_API_KEY=pgn_live_… openclaw config set channels.pidgeon '{}' --strict-json05 · Add the tools too
The channel answers the mail in front of the agent. The MCP server gives it the rest: new addresses, search, and waiting for a verification code.
openclaw mcp add pidgeon --url https://pidgeon.ai/api/mcp \ --transport streamable-http --auth oauth openclaw mcp login pidgeon
02 · Questions
Pidgeon in OpenClaw
- Do I need an API key?
- Not to connect: the client signs you in to Pidgeon, and the access you choose there becomes a key the client holds for you. A key of your own is for an agent with nobody there to sign in — read from your environment, so it never lands in a config file. Either can be scoped to one address and one level of access.
- How do I disconnect a client?
- Under Developers → API keys → Connected apps. Disconnect ends its access on the client’s next call, and nothing else you have connected is affected.
- Can my agent sign itself up?
- Yes. Paste “Read pidgeon.ai/skill.md and get yourself an email address.” into the agent. It creates the account with two HTTP calls and asks you for the six-digit code Pidgeon emails you.
- What can the agent do once it is connected?
- Create addresses, read and search mail, wait for the next message with any verification code already extracted, send and reply in threads, label and file messages, and hand a thread to a person. Every inbound message carries a Pidgeon Firewall verdict.
- Does new mail wake my OpenClaw agent?
- With the channel, yes: an email to the address starts a conversation within seconds, one per email thread, and the answer goes back as a reply in that thread. The MCP server alone gives the agent tools but does not wake it.
- What happens to a prompt-injection email?
- Mail Pidgeon Firewall marks suspicious — addressed to an AI, or asking to send mail somewhere else — is held in the inbox and never reaches the agent, and openclaw channels status counts it. The agent reads the Firewall’s clean text, never raw HTML, and can only reply in the thread, so one email cannot make it write to anyone else.
- What if my Gateway is down when mail arrives?
- It is answered when the Gateway is back. The channel reads from a cursor kept in OpenClaw’s state and queues each message durably before moving on, so a restart or a dropped connection resumes where it stopped.
03 · Elsewhere