Connect any AI agent to your help desk
Deskwoot speaks the Model Context Protocol, so an OpenClaw bot or any MCP client can work your real queue: read, reply, draft for approval, manage contacts, and start conversations. With scoped tokens you control and an audit trail behind every action.
Works with OpenClawScoped bot tokensEnterprise and free trial
{
"mcpServers": {
"deskwoot": {
"url": "https://app.deskwoot.com/api/mcp",
"headers": {
"Authorization": "Bearer dw_mcp_your_token"
}
}
}
}
A standard way to hand an AI agent the tools to do the job
The Model Context Protocol is how a modern AI agent gets a defined set of tools it can call. Instead of writing a brittle one-off integration and babysitting it, you connect an MCP client and the agent discovers exactly what it is allowed to do. For support that is the difference between a bot that drafts text in a vacuum and an agent that works inside your real help desk.
What your agent can actually do
A focused set of tools, not a master key. The agent calls tools/list, sees exactly these, and gets to work.
Read and search conversations
List, search, and read full message history across every channel the token can see.
Reply to customers
Send a real reply on the original channel, attributed to the agent in the thread.
Leave private notes
Drop context only your team sees, so a human picks up with everything they need.
Draft replies for approval
Propose an answer that stays pending until a human approves or edits it. Nothing ships unsupervised unless you allow it.
Manage contacts
Find, create, and update contacts, set custom attributes, and add notes to a profile.
Start outbound conversations
Open a new thread to a customer, the tool that makes onboarding nudges, follow-ups, and outreach possible.
Set status, assign, label
Move a conversation through your workflow: status, priority, labels, attributes, and escalation to a person.
Poll the events feed
Ask what changed since the agent last looked, so it can react to new messages and updates.
A teammate with a name badge and a leash
Letting an AI agent into your inbox should not mean handing over the keys. Here is what keeps it in bounds.
Scoped tokens
Each token gets only the permissions you grant: read only, reply, manage contacts. No master key.
Restricted to inboxes
Lock a token to specific inboxes, so an agent only ever touches the queues you point it at.
Revocable instantly
Pull a token and the agent loses access on the spot. No waiting, no redeploy.
Every action audited
The agent acts as its own service account, so every reply, note, and status change is logged to it.
Locked to your workspace
Tenant isolation holds at every layer. There is no path to another company's data.
Idempotent writes
Outbound sends carry an idempotency key, so a retry never double-posts to a customer.
What the agent can never touch
- Change account settings or channel config
- Touch billing or payment methods
- Delete conversations or contacts
- Send mass broadcasts
- Manage webhooks or export data
Three steps, no backend to host
- 1
Enable the MCP server
In Deskwoot, open Settings, then MCP Server, and switch it on. The first time provisions the agent's service account.
- 2
Mint a scoped token
Create a bot token, pick its scopes, and restrict it to the inboxes it should touch. You see the raw token once.
- 3
Point your agent at the URL
Add the connection URL and the token as a bearer header to your OpenClaw config. The agent discovers the tools and starts working.
{
"mcpServers": {
"deskwoot": {
"url": "https://app.deskwoot.com/api/mcp",
"headers": {
"Authorization": "Bearer dw_mcp_your_token"
}
}
}
}One connection, inbound and outbound
Inbound support triage
The agent reads incoming conversations, answers the routine ones from your help center, drafts replies for the tricky ones, and escalates anything about money or a frustrated customer to a person.
Proactive outreach
Because it can start outbound conversations, the agent runs onboarding check-ins, follows up on stalled tickets, and handles outreach to creators and partners, all from your real help desk.
Around-the-clock cover
An MCP agent works the queue when your team is offline, drafting and holding replies for morning review or handling the simple stuff on its own, within the limits you set.
Frequently asked questions
What is an MCP server for customer support?
An MCP (Model Context Protocol) server exposes your help desk to an AI agent as a defined set of tools. Instead of a custom integration, the agent connects once and can read conversations, reply, draft answers, and manage contacts, within the scopes you grant.
Can I connect an OpenClaw bot to Deskwoot?
Yes. OpenClaw and any other MCP-aware client connect the same way: enable the MCP server, mint a scoped bot token, and point the agent at app.deskwoot.com/api/mcp with the token as a bearer credential. The agent then discovers the available tools and starts working.
What can the AI agent NOT do?
The MCP surface excludes account settings, billing, deleting data, mass broadcasts, and webhook management. Those stay with humans in the dashboard. The agent can work conversations, drafts, notes, and contacts, and nothing more.
Is it safe to give an AI agent access to my inbox?
Tokens are scoped, can be restricted to specific inboxes, and are revocable instantly. Every action is attributed to the agent's service account in the audit log, and access is locked to your workspace with no path to another company's data.
Do I need an Enterprise plan to use MCP?
The MCP server is an Enterprise feature, and it is also available during the 7-day Enterprise trial that every new Deskwoot account starts on, so you can try it before you buy.
How does the agent know when something changes?
It polls an events feed and asks what has changed since it last looked. That keeps it reacting to new messages and updates without you having to wire up a separate webhook listener.