AnyHook

anyhook-mcp

An MCP server that gives AI agents eyes on webhooks, inspect events, replay deliveries, mock signed payloads, create apps. Designed for the 45-second AI handler workflow.

Apache 2.0 · npm i -g anyhook-mcp

Forward webhooks to localhost

A public webhook URL that forwards to your machine, with no account:

npx -y anyhook-mcp listen 3000

Webhook URL   https://in.anyhook.net/q-3fa9c1d2/listenForwards to   http://localhost:3000/
This URL is temporary and lasts until Nov 11. To keep it: https://anyhook.net/claim/...
Paste the URL into Stripe, GitHub or any webhook sender. Ctrl-C to stop.While this is stopped, events are logged but not forwarded.
Connected. Waiting for events.12:00:03  200  41 ms  invoice.paid12:00:19  500  12 ms  invoice.paid  AnyHook retries on its schedule

• The URL stays. Restart the command and it is the same URL, so there is nothing to paste into the provider again.
• Signatures verify. Your server gets the bytes the provider sent, so stripe.webhooks.constructEvent and friends work as they do in production.
• Nothing is lost when your laptop is shut. Events are stored either way. A failed delivery is retried, and one that arrived while this was stopped can be replayed.

listen 3000/api/webhooks sets the path. With ANYHOOK_API_KEY set it uses that account, and --app <slug> picks the app. Needs Node 22 or newer.

It is for webhooks, not a general tunnel. The sender gets its answer from AnyHook, so your server's response goes to the event log rather than back to the sender. To expose a whole site, use ngrok or Cloudflare Tunnel.

Already have an AnyHook account?

Connect the hosted endpoint instead of running anything: add https://anyhook.net/mcp to Claude Code, claude.ai, Codex or VS Code and sign in through the browser, no key to copy. Steps for each app: https://anyhook.net/mcp.

No account yet (local mode)

No API key needed to start:

claude mcp add anyhook -- npx -y anyhook-mcp

Then ask your agent to run anyhook_quickstart, it creates a free relay endpoint + API key instantly (no signup), and this MCP session auto-connects. All account tools (events, replay, apps) work immediately. The response includes a claim_url to keep the endpoint permanently; unclaimed, it expires in 7 days. It is a new endpoint, not an account you already have.

What it does

Drop this MCP server into Claude Desktop / Cursor / Claude Code and your agent can:

• anyhook_apps_list, see every app in your AnyHook account with its inbound URL
• anyhook_apps_create, spin up a new app from a prompt
• anyhook_connect, give a Stripe key or GitHub token: the webhook is registered at the provider and signature-checked from the first event (the key is not stored)
• anyhook_events, list recent events, filter by app / status
• anyhook_inspect, pull a single event's headers, body, signature status
• anyhook_replay, re-send a stored event to its destinations (outside the monthly quota, within a daily replay limit and the compute allowance; an event held for quota counts as new)
• anyhook_undelivered, list events that never reached a destination
• anyhook_usage, the plan's limits and what the account has used of each, with when they reset
• anyhook_replay_failed, bulk-replay an app's failed events, oldest first (the last 24 hours and 50 per call by default, up to 100), saying when more are waiting
• anyhook_mock, generate a signed Stripe / GitHub / Slack payload for local handler testing
• anyhook_verify, verify any incoming signature against your secret
• anyhook_providers, list every webhook provider AnyHook supports

The inbound URL an app gives you accepts webhook POSTs and also answers provider setup handshakes automatically: Meta's GET carrying hub.mode=subscribe gets its hub.challenge echoed back, even before the app is configured, so the URL can be registered in the Meta App Dashboard first and wired up after.

Two modes:

Quick start

npm install -g anyhook-mcp# or run without installingnpx -y anyhook-mcp

Get an API key from https://anyhook.net/dashboard/settings. Keys start with ahk_live_.

Install in Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{  "mcpServers": {    "anyhook": {      "command": "npx",      "args": ["-y", "anyhook-mcp"],      "env": {        "ANYHOOK_API_KEY": "ahk_live_xxxxxxxx"      }    }  }}

Restart Claude Desktop. The tools appear under the hammer icon.

Install in Cursor

Open Settings → MCP → Add new MCP server. Or edit ~/.cursor/mcp.json:

{  "mcpServers": {    "anyhook": {      "command": "npx",      "args": ["-y", "anyhook-mcp"],      "env": {        "ANYHOOK_API_KEY": "ahk_live_xxxxxxxx"      }    }  }}

Install in Claude Code

claude mcp add anyhook -e ANYHOOK_API_KEY=ahk_live_xxxxxxxx -- npx -y anyhook-mcp

Or add the same mcpServers entry to your project's .mcp.json.

Remote endpoint (no local process)

The same server is hosted at https://anyhook.net/mcp (streamable HTTP, stateless). Claude Code, claude.ai, Codex and VS Code sign in through the browser (OAuth), no key to copy: https://anyhook.net/mcp has the steps for each. Any other client sends Authorization: Bearer ahk_live_... with a key from the dashboard:

{ "mcpServers": { "anyhook": { "url": "https://anyhook.net/mcp", "headers": { "Authorization": "Bearer ahk_live_..." } } } }

The endpoint has no anyhook_quickstart: that tool exists only in local mode.

Local-only mode (no account needed)

Skip the env var:

{  "mcpServers": {    "anyhook": {      "command": "npx",      "args": ["-y", "anyhook-mcp"]    }  }}

The agent gets anyhook_mock, anyhook_verify, anyhook_providers, anyhook_simulate, anyhook_events (against memory store), and anyhook_inspect (against memory store). Useful for "generate me a signed Stripe payment_intent.succeeded and POST it to localhost:3000" flows during local dev.

Example conversations

Triage a failing endpoint:

"Show me undelivered events for the stripe-prod app from the last hour, then replay them."

The agent calls anyhook_undelivered then anyhook_replay for each, surfacing the responses inline.

Create an app for a new integration:

"Make a new AnyHook app called replicate-video-jobs, source generic, point it at https://my-app.vercel.app/api/replicate-callback."

Calls anyhook_apps_create, returns the inbound URL you paste into Replicate.

Sanity-check a signature failure:

"Here's an inbound payload from Stripe that AnyHook is rejecting. Verify the signature manually with secret whsec_xxx."

Calls anyhook_verify with the supplied headers + body + secret, returns valid: false plus the failure reason.

Mock a webhook for local testing:

"Generate a Stripe payment_intent.succeeded event and POST it to http://localhost:3000/api/webhooks/stripe."

Calls anyhook_mock with targetUrl set; returns the request that was sent + your handler's response.

Environment variables

Why this is interesting

AI agents have been blind to their own webhook infrastructure. They can write the integration code, but once an event misbehaves in production they can't see it, you tell them what happened. With this MCP server, Claude / Cursor can read the actual event log, replay deliveries, debug signature failures, and even spin up new apps. It closes the loop between writing webhook code and operating it.

License

Apache-2.0 © AnyHook