Use it from an AI agent (MCP)

testhooks mcp is a small MCP server that lets an AI coding agent use Testhooks the way you use the UI. It gives you a better handler, written from what the provider really sends instead of what its docs say it sends.

A typical session:

1. You ask the agent to “set up a webhook endpoint for Stripe”. It creates one and gives you the URL.
2. You paste that URL into the provider’s webhook settings and trigger an event.
3. The agent reads the requests that arrived (headers, body, signature) and writes the handler to match.
4. It replays a captured request to check the handler, as if the provider sent it again.

The MCP server runs on your machine and talks to a running Testhooks over the REST API. It adds nothing of its own: every tool is an API call you could make with curl.

Set it up

You need the testhooks binary on your machine (see Installation) and a Testhooks to talk to: your own (testhooks in another terminal), a deployed one, or the live demo.

Variable
TESTHOOKS_URL	optional	Address of the Testhooks to use. Defaults to http://localhost:8080
TESTHOOKS_AUTH_TOKEN	optional	The server’s AUTH_TOKEN, when it has one. Sent as Authorization: Bearer <token>

The capture URLs the agent hands out start with TESTHOOKS_URL, so a provider on the internet can only reach them when that address is public.

Claude Code

claude mcp add testhooks -e TESTHOOKS_URL=http://localhost:8080 -- testhooks mcp

Add -e TESTHOOKS_AUTH_TOKEN=<token> when the server needs one.

Cursor and Claude Desktop

Add this to .cursor/mcp.json (Cursor) or claude_desktop_config.json (Claude Desktop):

{
  "mcpServers": {
    "testhooks": {
      "command": "testhooks",
      "args": ["mcp"],
      "env": {
        "TESTHOOKS_URL": "http://localhost:8080",
        "TESTHOOKS_AUTH_TOKEN": ""
      }
    }
  }
}

If the app can’t find testhooks, use its full path, for example /usr/local/bin/testhooks.

Tools

Tool	What it does
create_endpoint	Creates a webhook URL that records every request sent to it. Takes an optional name, mode (server or browser) and config. Returns the endpoint with its capture_url
list_endpoints	Lists endpoints, each with its capture_url
get_endpoint	Shows one endpoint by id, or by the slug in its capture URL
update_endpoint_config	Changes only the settings you pass and keeps the rest. Set a field to null to remove it
list_requests	Lists the requests an endpoint received, newest first
get_request	Shows one request in full: method, path, query, headers, body, signature check, what was forwarded and what the sender got back
replay_request	Sends a captured request to its endpoint again with the same method, path, query, headers and body, and returns the reply. Connection headers such as Host and Content-Length are not copied
delete_endpoint	Deletes an endpoint and all its requests. This cannot be undone; it is marked destructive, and most MCP clients ask before running such tools

A replayed request is a new request: it shows up in the list, runs through the endpoint’s transform, forwarding and reply like any other, and the agent sees the status the provider would have seen.

Signature secrets stay write-only here too: update_endpoint_config keeps the stored secret unless you pass a new signature.secret.