The botmail prompt with your agent

A personal assistant that runs on your computer or server and answers you on Telegram, WhatsApp, Discord, Slack and other channels.

1. Store the token from the terminal with openclaw secrets store set BOTMAIL_TOKEN --kind env, or in ~/.openclaw/.env. That way it reaches the bot's commands without going through the chat.
2. Paste the prompt, without the Token line, into a direct chat with your OpenClaw.
3. Create the recurring check with openclaw automations create "*/10 * * * *" "Check the botmail inbox" --session isolated, or type /loop 10m Check the botmail inbox in chat.
4. OpenClaw calls botmail with curl through its exec tool. If you turned on command approvals, allow curl to botmail.it, or it will ask you at every check.

The web_fetch tool isn't enough, because it only sends GET requests without headers. If you store the token as a secret rather than a variable, turn on the secrets egress proxy and add botmail.it to the allowed hosts.

An agent with memory and skills it learns on its own. Runs on your computer or server and answers from the terminal, Telegram, Discord, Slack, WhatsApp and more.

1. Store the token on the machine running Hermes with hermes config set BOTMAIL_TOKEN followed by the token. It goes into ~/.hermes/.env.
2. In ~/.hermes/config.yaml, add BOTMAIL_TOKEN to terminal.env_passthrough.
3. Paste the prompt into the chat, without the Token line.
4. Schedule the check with /cron add "every 10m" "Check the botmail inbox. If there are no new messages reply [SILENT]". With [SILENT] you get no notification when there is nothing new. The gateway must be running (hermes gateway install).

Step 2 matters. Hermes strips every variable ending in _TOKEN from shell commands, and without env_passthrough the bot gets error 401.

A cloud assistant that works with routines and its own credentials.

1. Store the token as a credential named BOTMAIL_TOKEN. Lindy's credentials are write-only and reach the bot's commands and code.
2. Paste the prompt without the Token line.
3. Create a Routine on a schedule, for example every 15 minutes.

A cloud agent that runs long and scheduled tasks.

1. Add botmail.it as a custom API in the connectors, with the token as the key.
2. Paste the prompt without the Token line.
3. Create a schedule for the inbox check.

Manus's documentation on where it keeps keys is mostly about the apps it builds. Before you start, check that the token doesn't end up in the task text.

The calls to botmail are made by ChatGPT Work, which runs code in a cloud environment.

1. In Settings › Data controls › Work network access, allow public internet access.
2. Store the token as a network secret, with key BOTMAIL_TOKEN and domain botmail.it. ChatGPT adds it to calls to botmail without showing it to the model.
3. Paste the prompt into a Work conversation, without the Token line.
4. Create a scheduled task every 10 or 15 minutes that checks for unread messages.

Don't build a custom GPT with Actions: OpenAI has stopped new GPTs and switches them off on 11 December 2026. On Plus, Work is still rolling out.

claude.ai, Claude Desktop, the mobile app and Cowork connect through a connector: they use no prompt, and the token never leaves the dashboard. Claude Code, in the terminal or the desktop app, uses the prompt as is. Cloud routines need two settings.

1. claude.ai, Claude Desktop, the mobile app and Cowork: in claude.ai open Customize › Connectors › Add custom connector, enter https://botmail.it/mcp, keep the default sign-in options and click Add, then Connect. In the botmail window, sign in with your login code if asked, and click Authorize.
2. Don't paste the prompt into these apps. Claude can't add a connector by itself, and the token would stay in the chat history. Once connected, Claude already has botmail's tools: ask it in plain words what to do with the mail, or set up a scheduled task, for example “Every hour, read my botmail mail and summarise the invoices”.
3. Claude Code: put the token in an environment variable, for example in the env key of the project's local settings. Never in CLAUDE.md. Then paste the prompt without the Token line, or put it in the project's CLAUDE.md. For MCP, add the server with claude mcp add and use the MCP prompt: the MCP clients card below has the details.
4. Claude Code without a token: claude mcp add --transport http botmail https://botmail.it/mcp without --header, then /mcp to sign in to botmail and authorize. As with connectors, no prompt is needed: give Claude the task in plain words.
5. For Claude Code's checks, use the desktop app's scheduled tasks, which need the computer on. /loop 10m only runs while the session is open and expires after 7 days.
6. Cloud routines: the default network blocks botmail.it. Store the token as a network secret (Authorization header, Bearer prefix, host botmail.it), or add botmail.it to the environment's network. Routines run at most once an hour.

We verified the claude.ai connector in October 2026. It signs in to botmail with OAuth: you copy no token, and you can revoke the connection on the token page. In the claude.ai chat, code can't reach arbitrary domains, so the HTTP API prompt doesn't work there.

The prompt works with Gemini CLI. The Gemini app has no documented way to call an arbitrary API with a token.

1. Install Gemini CLI on a computer or server that stays on.
2. Put the token in .gemini/.env and the prompt, without the Token line, in GEMINI.md.
3. Gemini CLI has no scheduler. Use cron, for example */10 * * * * gemini -p "Check the botmail inbox" --yolo.

With --yolo Gemini approves every command by itself, so run it in a dedicated folder. The Gemini Spark agent isn't available in the EEA, the UK or Switzerland.

You use an autonomous Copilot Studio agent with an HTTP node. Copilot chat doesn't call outside APIs.

1. Create an autonomous agent with a recurrence trigger every 10 minutes.
2. Store the token as a secret environment variable, backed by Azure Key Vault.
3. Add an HTTP Request node: GET https://botmail.it/api/v1/messages?unread=true with an Authorization header set to Bearer followed by the token.
4. Copy the prompt's safety rules into the agent's instructions.

The HTTP node gives up after 30 seconds, so use the unread list, not the /messages/wait call.

The Grok app reaches outside services only through MCP connectors, and its automations run at most once a day. botmail now offers OAuth sign-in for connectors. What we have not confirmed is that Grok speaks protocol version 2026-07-28.

1. Try adding https://botmail.it/mcp as a connector and signing in to botmail when it asks.
2. If connecting fails, use a Grok model through the xAI API inside a tool on this page, like n8n or your own script, which makes the calls to botmail.

ChatGPT's always-on agents, introduced at the end of September 2026.

We found no documented way for a dot to call an API with a token. Use ChatGPT Work, above.

On the Pro plan, dots aren't available in the EEA, Switzerland or the UK.

Perplexity Computer reaches outside services through MCP connectors, and its tasks run at most once an hour. botmail now offers OAuth sign-in for connectors. What we have not confirmed is that Perplexity speaks protocol version 2026-07-28.

1. Try adding https://botmail.it/mcp as a connector and signing in to botmail when it asks.
2. If connecting fails, connect Perplexity to botmail through an automation like n8n, or pick an agent from the first section.

Automated flows built from nodes, with an AI agent inside if you want one.

1. Create a credential of type Bearer Auth with your token.
2. A Schedule Trigger every 5 or 10 minutes, then an HTTP Request node: GET https://botmail.it/api/v1/messages?unread=true with that credential.
3. For each message: GET /messages/{id} to read it, your own logic, then DELETE /messages/{id}.
4. Want an agent? Attach the HTTP Request node as a tool to the AI Agent node and put the prompt's rules in its system message.

Publish the workflow, or the schedule won't run.

Connects botmail to thousands of apps.

1. Trigger Webhooks by Zapier, event Retrieve Poll, URL https://botmail.it/api/v1/messages?unread=true, Authorization header set to Bearer followed by the token.
2. In the Key field, write messages. Zapier uses each message's id so it never handles one twice.
3. Add your actions. To delete the message, use Webhooks by Zapier › Custom Request with method DELETE.

Webhooks by Zapier needs a paid plan, Professional or above. Zapier checks every 2 minutes on Professional and every minute on Team.

Automated scenarios built from modules.

1. HTTP module › Make a request: GET https://botmail.it/api/v1/messages?unread=true.
2. Put the token in the module's credentials, an API key keychain on the Authorization header, with the value Bearer followed by the token.
3. Schedule the scenario: every 15 minutes on the free plan, down to every minute on paid plans.

Automated flows in Microsoft 365.

1. A Recurrence trigger every 5 or 10 minutes.
2. An HTTP action: GET https://botmail.it/api/v1/messages?unread=true with the Authorization header.

The HTTP action is a premium connector and needs a Premium licence.

OpenAI's agent for the terminal and the cloud.

1. Put the prompt, without the Token line, in AGENTS.md.
2. In the CLI the network is off. Turn it on with network_access = true in the sandbox_workspace_write section of the config.
3. In the cloud, choose internet access for custom domains only, add botmail.it and store the token as a network secret.
4. Schedule with the app's scheduled tasks, or with cron and codex exec.

Classic Codex Cloud secrets exist only during the setup script and are gone before the agent starts. Use network secrets.

If you build your own agent, botmail is four tools.

1. Write one tool per call (wait, list, read, delete) that reads the token from BOTMAIL_TOKEN.
2. Copy the prompt's safety rules into the agent's instructions.
3. Schedule with cron or the scheduler you already use.

If the library speaks MCP 2026-07-28, it can use botmail's MCP server directly instead of the four tools. The reference "fetch" MCP server won't do: it allows no custom headers and no DELETE.

The server is `https://botmail.it/mcp` and speaks only protocol version 2026-07-28. Connector apps, such as claude.ai, connect with OAuth sign-in and use no prompt: the token never leaves the dashboard. Apps with a configuration file use the MCP prompt, next to the HTTP API prompt, and the token in a header.

1. Connector apps, such as claude.ai: add https://botmail.it/mcp as a custom connector and sign in to botmail when it asks. You approve the connection and copy no token. Don’t paste the prompt: once connected, the app already has the tools, and you tell it in plain words what to do with the mail.
2. Claude Code: claude mcp add --transport http botmail https://botmail.it/mcp --header "Authorization: Bearer $BOTMAIL_TOKEN", with the token in the BOTMAIL_TOKEN variable. Or leave out --header, then run /mcp to sign in to botmail with OAuth like a connector.
3. Clients with a JSON configuration file: a server with url https://botmail.it/mcp and an Authorization header set to Bearer followed by the token.
4. Your own agent: the official MCP TypeScript SDK, version 2, with version negotiation on, or the OpenAI Agents SDK for JavaScript.
5. With the token in a header, paste the MCP prompt without the Token line: the client already has the token.

We verified claude.ai connectors in October 2026. For Cursor, Codex and ChatGPT we found no confirmation of version 2026-07-28. ChatGPT should now be able to connect with OAuth, but we have not verified it yet. VS Code still speaks only the 2025 versions and will not connect. If connecting fails, use the HTTP API prompt.