Connect an AI agent with the MCP server (Enterprise)
The MCP server lets an external AI agent operate your helpdesk through a scoped bot token. It is a Model Context Protocol endpoint, so any MCP-aware agent (for example an OpenClaw bot) can read conversations, reply to them, leave private notes, draft replies for a human to approve, manage contacts, start new outbound conversations, and poll an events feed. You enable it under Settings, mint a token, and point your agent at the connection URL. The MCP server is available on the Enterprise plan and during the 7-day Enterprise trial that every new account starts on.
What is the MCP server for?
It is the bridge between an outside AI agent and Deskwoot. Instead of writing your own integration against the REST API, you connect an MCP client and the agent gets a defined set of tools it can call. Think of it as giving a bot a scoped, revocable key to a specific part of your inbox rather than handing over a login. The agent acts through a bot identity, so its messages and notes are attributed to that bot, not to a human teammate.
How do I enable the MCP server?
- Go to Settings > MCP Server.
- Turn the MCP server on for your workspace.
- Mint a bot token. Give it a clear name (for example OpenClaw triage bot) so you can recognize it later.
- Choose the scopes the token should have, and limit it to specific inboxes if the agent only needs to work in some of them.
- Copy the token right away. For security it may not be shown in full again after you leave the page.
What is the connection URL?
Point your MCP client at:
https://<your-host>/api/mcp
Replace <your-host> with your own Deskwoot host. Pass the bot token the way your MCP client expects (usually a bearer token in the authorization header). Once the client connects, the agent will see the tools it is allowed to use based on the token's scopes.
What can an external agent do through MCP?
A connected agent, within the scopes you grant, can:
- Read conversations and their messages.
- Reply to conversations.
- Add private notes that only your team sees.
- Create draft replies that a human teammate reviews and sends.
- Read, create, and update contacts.
- Start outbound conversations.
- Poll an events feed to react to new activity.
What can it not do?
The MCP server is deliberately narrow. It does not expose your workspace settings, billing, or any destructive action. Specifically, a bot token cannot change settings or configuration, touch billing, delete records, send broadcasts, or manage webhooks. If your agent needs to do something outside that boundary, the MCP server will not do it, and that is on purpose.
How do scopes and inbox limits work?
Each token carries a set of scopes that decide which tools the agent can call. Grant only what the agent actually needs. A bot that drafts replies for human approval does not need permission to send messages on its own, for instance. On top of scopes, you can restrict a token to specific inboxes, so an agent that handles your billing inbox never sees conversations in another one. If a token is ever leaked or you stop using an agent, revoke it under Settings > MCP Server and the access is gone immediately.
Which plans include the MCP server?
The MCP server is an Enterprise feature. It is also available during the 7-day Enterprise trial, so you can wire up and test an agent before you commit to a paid plan. If you need programmatic access without MCP, the REST API covers the same data and is also an Enterprise feature, with tokens managed under Settings > API tokens.