Customer support for MCP servers

Your users are in Claude now. Put your help desk there.

canhelpto mounts support inside your MCP server: three tools any agent can call, the failure context attached by your server, a human inbox, and the answer back in the conversation. The same desk serves your web users through a widget, email and WhatsApp.

Start freeRead the docs for agents

Install with one prompt to your coding agent: Add canhelpto support to this MCP server, following canhelpto.com/install.md

When a customer uses your product through an agent, your web widget is never on screen. The user asks the agent why it fails, and the agent has nothing to call. Tickets arrive as "it doesn't work", if they arrive at all.

How it works

Mount, deflect, hand off. Nothing leaves your server without your rules first.

Step 1

Mount

Drop the wrapper on your MCP server, or call three REST endpoints from any agent framework. Ten minutes, one workspace key.

Step 2

Deflect

When a tool fails, the error already names support_ask. The agent gets an answer it can act on: use this parameter, reconnect here, wait this long.

Step 3

Hand off

What the rules cannot settle becomes a ticket with the context attached. Your team answers from the inbox; the reply goes back by email and through the tool.

Three tools, in your server

Three entries in your tool list. No extra connector for the user to install.

support_ask

Answers from your own rules, in code, with no model call. Lists the account's recent errors. An unknown question is never invented; it points at the ticket.

support_open_ticket

Opens a ticket in your inbox with the customer's words plus the context your server attaches: account, client, catalogue version, the last tool calls with their errors.

support_ticket_status

Reads the ticket and your team's replies back into the conversation. Internal notes stay out. Ownership is checked against the account.

The ticket arrives with the evidence

Your server already knows who called, from which client, and which call failed with which error. canhelpto writes that under the customer's words, so your team reads the cause before the complaint.

  • Ownership checked in your server: a customer only ever reads their own ticket.
  • Agent agnostic: MCP tools for Claude, ChatGPT, Cursor and your own agent; REST for everything else.
  • One inbox for the agent channel and the web: widget, email, WhatsApp, feedback, CSAT.
Claude cannot approve my drafts, it says the account is managed.

- Connector context (attached by the vendor's MCP server) -
Account: Ana Pop · [email protected] · user 7f3e… · user · managed
Client: claude-user/1.0 · catalogue 1.8.0 · connector added 2026-09-12
Last calls (newest first):
- 16:05:02 list_drafts · 340 ms · ok
- 16:05:00 approve_drafts · 120 ms · FORBIDDEN: Your account is managed by the team

Free to start. EUR 49 per server when the AI answers for you.

Unlimited tickets on Free. Pro adds AI answers from your knowledge base behind a confidence gate, analytics per tool, and secret keys.

See pricing

Questions

What is canhelpto?
canhelpto is customer support for MCP servers. A software vendor mounts three tools inside its own MCP server so a user working in Claude, ChatGPT or Cursor can ask for help, open a ticket that carries the failure context, and read the team's answer back, without leaving the conversation.
How is this different from a docs bot or a help desk with its own MCP server?
A docs bot answers from documentation and opens no ticket. A help desk with its own MCP server serves your team, or asks the end user to install a second connector. canhelpto lives inside the vendor's own server, so the ticket arrives with the account and the last failed calls, and a human can take over.
Which agents does it work with?
Any MCP client (Claude, ChatGPT, Cursor, VS Code, your own agent) through the three tools, and any other agent through plain REST. The desk is the same one your web users reach through the widget, email and WhatsApp.
Does the AI answer customers on its own?
On Free, the rules answer in code and AI drafts a reply that a human sends. On Pro, AI answers from your knowledge base when its confidence clears a gate you set; below it, a human takes the ticket.
What does it cost?
Free for one workspace with unlimited tickets. Pro is EUR 49 per MCP server per month. The first three vendors run Pro at no cost through October 2026.