Webhook debugger for coding agents

Debug webhooks by asking your coding agent

Connect Claude Code, Cursor, or Codex over MCP or the CLI. Your agent gets 46 tools to inspect payloads, replay requests, and verify the fix while you watch.

Start free

No credit card, see pricing. The live endpoint is already listening.

live eventsconnecting
# creating your ephemeral endpoint...

The test event fires a Stripe-shaped payload at your endpoint. No terminal, no signup.

Trusted by engineers at

  • Shopify
  • Docker
  • Boeing
  • Cisco
  • IBM
  • T-Mobile
  • Okta
  • Wiz
  • Razorpay
  • HackerOne

Five tools, two operators

Five webhook tools that answer to you, and to your agent.

Hooklistener is a webhook debugger with an MCP server. You work in the dashboard; your coding agent works over MCP or the CLI. Same endpoints, same captured requests, same replay — nothing here is dashboard-only.

claude code

Show me the last Stripe webhook payload

MCPget_request
"type": "invoice.paid",
"amount": 4999,
"customer": "cus_R8x..."

46 MCP tools available

Endpoints, requests, monitors, email, automations, diagnosis

MCP server. Hand webhook debugging to your coding agent

Connect Claude Code, Cursor, or Codex to Hooklistener over MCP. Your agent creates the endpoint, waits for the webhook to land, reads the payload, and tells you what broke — without you opening a browser tab.

Drive it fromClaude Code · Cursor · Codex · Any MCP client

Learn about the MCP server
  • 46 tools across 9 categories, endpoints through AI diagnosis
  • Works with Claude Code, Cursor, Codex, any MCP client
  • Streamable HTTP transport (JSON-RPC 2.0), OAuth 2.1 + PKCE
  • One command to install, no config file to hand-edit
Secure by default. TLS on every endpoint, encrypted storage, custom response codes that keep your internals private, and role-based access on team plans.

MCP server

Your agent debugs the webhook over MCP. You read the summary.

Connect once and Claude Code, Cursor, or Codex get 46 tools for endpoints, events, replays, monitors, and email inboxes, then use them without leaving the editor.

claude code — hooklistener mcp46 tools connected
  1. You:

    Stripe checkout webhooks are failing in staging. Figure out why.

  2. Agent calls the Hooklistener MCP tool
    MCPlist_requests(endpoint: "stripe-staging", status: "5xx", since: "2h")0.4s
    14 failures · all checkout.session.completed · first at 09:12
  3. Agent calls the Hooklistener MCP tool
    MCPget_request(id: "evt_9f2ac41")0.3s
    500 from your app — signature rejected
    "type": "checkout.session.completed"
    "stripe-signature": "t=1754382712,v1=8b3c…"
    "response_status": 500
    "response_body": "Webhook signature verification failed"
  4. Agent:

    The signature header is valid on all 14 — staging is the one rejecting them. Your signing secret was rotated at 09:07, five minutes before the first 500. STRIPE_WEBHOOK_SECRET still holds the old value. Updated it and redeployed.

  5. Agent calls the Hooklistener MCP tool
    MCPforward_request(endpoint: "stripe-staging", ids: "14 failed")2.1s
    14 replayed · 14 × 200 · avg 138 ms
  6. Agent calls the Hooklistener MCP tool
    MCPcreate_monitor(endpoint: "stripe-staging", silent_for: "30m", notify: "#payments")0.2s
    monitor active
  7. Agent concludes:

    Root cause: a rotated signing secret staging never picked up. All 14 checkout events replayed clean, and the next silent half hour pings #payments before a customer does.

Set up the MCP server

One command, no config. Prefer a terminal? The same tools ship in the CLI.

CLI tunnel

Tunnel webhooks to localhost with one command.

Get a public HTTPS URL that forwards straight to your dev server. Every request through the tunnel is captured, inspectable, and replayable — and your coding agent can open one itself, since the tunnel is just a CLI command.

Compare with ngrok

Custom responses

Make your test endpoint fail on purpose.

Providers time out, rate-limit you, and return shapes their docs never mention. Set the status, headers, body, and latency your endpoint replies with, and reproduce that behavior whenever you need it instead of waiting for it in production.

  1. Set the response

    Choose the HTTP status, add response headers, and write the JSON or plain-text body your endpoint returns to the caller.

  2. Add latency and errors

    Hold responses for a fixed delay or return intermittent 5xx errors, then let your agent run the integration against it and confirm the retry and backoff paths actually work.

  3. Reuse it everywhere

    The same endpoint works in local dev, CI, and staging. Change the response rules per environment without touching application code.

How this compares to webhook.site

Questions

Frequently asked questions about Hooklistener

Everything else lives in the documentation (opens in a new tab).

What is Hooklistener?
Hooklistener is a webhook debugger built for AI-assisted development. An MCP server lets Claude Code, Cursor, or Codex debug webhooks on your behalf. Debug Endpoints capture and inspect every incoming request. Monitors alert you when an integration goes silent. Email Inboxes turn incoming email into webhooks. And the CLI tunnels live payloads to localhost. Your agent works through MCP or the CLI, you work through the dashboard, and both see the same requests.
Can I use Hooklistener for free?
Yes, and no credit card is required. The Free plan includes 1 Debug Endpoint with 1 day of request history, 1 Email Inbox, 1 Monitor, and full MCP access, so you can connect your coding agent over OAuth on day one. When you need more, Pro adds unlimited requests, 10 endpoints, 14-day retention, and CLI tunneling. Sign up free or compare plans.
Can my AI coding agent use Hooklistener?
Yes. Hooklistener ships an MCP server, so Claude Code, Cursor, Codex CLI, and any client that speaks the Streamable HTTP transport can drive it directly. Setup is one command:claude mcp add --transport http hooklistener https://app.hooklistener.com/api/mcpOAuth signs you in through the browser, so there is no API key to paste. Your agent then has 46 tools covering Debug Endpoints, captured requests, automations, schedules, secrets, Email Inboxes, and Monitors. In practice that means it can create an endpoint, block until the next webhook lands, read the payload, and explain why your handler returned a 500, without you leaving the terminal. Setup guide for Claude Code, Cursor, and Codex.
What is the Hooklistener MCP server?
MCP (Model Context Protocol) is an open standard for exposing tools to AI assistants. The Hooklistener MCP server is a hosted endpoint at https://app.hooklistener.com/api/mcp that publishes 46 webhook tools across 9 categories: Debug Endpoints, requests, automations, schedules, secrets, a key-value datastore, Monitors, Email Inboxes, and AI request analysis. It runs over Streamable HTTP (JSON-RPC 2.0), authenticates with OAuth 2.0 (legacy hklst_ API keys still work), and is available on every plan including Free. There is nothing to install or self-host. See the full tool list or read what an MCP server is.
Do I still get a dashboard if my agent does the debugging?
Yes. The MCP server, the CLI, and the dashboard all read the same data. Your agent can create endpoints and inspect payloads from the terminal while you watch requests stream in live in the browser, filter them, replay them to any destination, and share a read-only inspection link with a teammate. Anything your agent does appears in the dashboard immediately, and anything you do is visible to the agent on its next call.
What are Debug Endpoints?
Debug Endpoints are unique HTTPS URLs that capture every incoming HTTP request with full details: headers, body, query params, IP address, and TLS information. You can inspect requests in real time, replay them to any destination, share inspection links with teammates, and keep up to 60 days of history on paid plans. They are built for debugging third-party integrations without touching your production code. See the webhook debugger.
How do Monitors work?
Monitors watch your Debug Endpoints for inactivity and send alerts when webhooks stop arriving. You configure a timeout threshold (1 to 1440 minutes), and if no requests arrive within that window, Hooklistener notifies you by email or Slack. Cooldown periods prevent alert spam, and you can track alert history and last-request timestamps to diagnose issues quickly. More on webhook monitoring.
How does the CLI work?
The Hooklistener CLI lets you receive live webhook payloads on your local machine. After authenticating with a device code flow, you can create a tunnel that forwards requests from any Debug Endpoint to localhost. It works behind firewalls, requires no ngrok setup, and gives you full request details including headers, body, and cookies for local debugging. CLI tunnel details.

One command to connect

Give your coding agent a webhook debugger.

Connect Claude Code, Cursor, or Codex to the MCP server and your agent gets 46 tools for endpoints, requests, automations, schedules, and email. You get the same data live in the dashboard.

Free plan includes MCP access. No credit card required.