openwebhook.coOpen inspector →
developer guide

Test Webhooks in Cursor and Claude with MCP

Connect the OpenWebhook MCP server to Cursor or Claude Code, list custom endpoints, watch a slug, and wait for the next webhook from the chat.

By David Lorenzo García··6 min read

The OpenWebhook MCP server lets an AI coding agent list your Pro endpoints, watch a slug, and block until the next HTTP request arrives. The same revocable token used by the CLI authenticates it. Nothing is written to stdout except JSON-RPC.

Install the public package:

npm install --global openwebhook

Tools

Tool Purpose
list_endpoints Slugs and IDs on the account
create_endpoint Create a custom slug
delete_endpoint Delete by endpoint ID
watch Start an in-memory SSE buffer (up to 100 events)
unwatch Stop the stream and drop the buffer
wait_for_webhook Block until the next event (timeout_ms, max 300000)

Payloads stay in the MCP process only. They are not written to disk.

Claude Code

claude mcp add openwebhook \
  --env OPENWEBHOOK_TOKEN=owh_live_YOUR_TOKEN \
  -- openwebhook mcp

Run /mcp and confirm openwebhook is connected. Then ask: “List my OpenWebhook endpoints.”

Cursor

Add a project or user MCP server that runs openwebhook mcp with OPENWEBHOOK_TOKEN. Restart the agent after saving.

If the GUI cannot find the binary, use the absolute path from which openwebhook, or:

{
  "mcpServers": {
    "openwebhook": {
      "command": "npx",
      "args": ["-y", "openwebhook", "mcp"],
      "env": {
        "OPENWEBHOOK_TOKEN": "owh_live_YOUR_TOKEN"
      }
    }
  }
}

A complete loop

  1. Ask the agent to watch your slug (for example billing).
  2. Ask it to wait_for_webhook with a 60 second timeout.
  3. In another terminal:
curl -X POST \
  'https://hooks.openwebhook.co/billing/probe' \
  -H 'content-type: application/json' \
  --data '{"from":"mcp"}'

The agent should return the method, path, headers and body. That proves the stream ticket, token, and tool wiring work.

When to use MCP versus the CLI

Use openwebhook listen --port when a real local HTTP server must receive the request. Use MCP when you want the payload inside the chat to write a handler, compare headers, or generate a test.

See Forward webhooks to localhost for the listen path.

Related guides