# Hookwave

> Security-first webhook gateway for AI / automation builders. Hookwave
> receives, verifies, transforms, and reliably delivers webhooks; lets
> developers replay past events without re-triggering the upstream
> provider; surfaces failures via an auto-grouping issues system; and
> exposes everything to AI agents through an MCP server.

Hookwave is the production webhook layer that sits between webhook
senders (Stripe, GitHub, Shopify, Replicate, custom senders, schedules)
and your code (HTTP endpoints, Slack, Discord, Teams, WhatsApp, n8n,
Make, Postgres, S3, Klaviyo, CLI tunnels). The pitch: verified
ingestion, fan-out routing, replay-without-re-triggering, and a
first-class AI/MCP surface so an LLM can list events, replay failures,
and diagnose problems from inside an editor.

## Concepts

- **Source**: an inbound webhook channel (Stripe, GitHub, Shopify,
  Replicate, generic, or a Hookwave-owned recurring schedule). Each
  source has an ingest URL and an HMAC verifier specific to its
  provider.
- **Destination**: an outbound delivery target — an HTTP endpoint, a
  Slack / Discord / Teams / Telegram incoming webhook, n8n / Make,
  Postgres / S3, Klaviyo, Twilio (SMS / WhatsApp), or a `mock` sink
  for testing.
- **Connection**: source → destination wiring with optional filters,
  transformations, retry policy, outbound auth, and headers. One
  source can fan out to many connections; settings live on the
  connection, not the destination.
- **Event**: a single inbound request that's been captured. Each event
  produces zero or more deliveries (one per matching connection).
- **Issue**: an auto-grouped delivery problem (consecutive failures,
  source silent, schema drift, status-code seen, verification
  failures) with status and resolution flow.
- **Trigger**: a rule that opens an issue when a condition is met
  (e.g. "5 consecutive failures within 10 min on connection X").

## Documentation

- [CLI reference](https://hookwave.dev/docs/cli): every command in the
  Hookwave CLI — `login`, `listen`, `init`, `events`, `doctor`,
  sources / destinations / connections CRUD, `tokens`, `mcp`.
- [MCP & Skills](https://hookwave.dev/docs/mcp): how AI assistants
  connect to Hookwave via the Model Context Protocol. Client configs
  for Claude Desktop, Cursor, Continue, and Claude Code, plus the
  full tool catalogue.
- [llms-full.txt](https://hookwave.dev/llms-full.txt): single-file
  concatenated copy of the CLI + MCP + concept docs, optimised for
  pulling into an LLM context window.

## Pricing & legal

- [Plans](https://hookwave.dev/plans): pricing tiers and feature
  comparison.
- [Status](https://hookwave.dev/status): live system status and
  recent incidents.
- [Security](https://hookwave.dev/security): security posture and
  vulnerability disclosure program.
- [Terms](https://hookwave.dev/terms), [Privacy](https://hookwave.dev/privacy),
  [DPA](https://hookwave.dev/dpa), [AUP](https://hookwave.dev/aup),
  [Subprocessors](https://hookwave.dev/subprocessors).

## MCP tools (for AI agents)

- `hookwave_whoami` — identify the calling org and plan.
- `hookwave_list_sources` / `hookwave_list_destinations` /
  `hookwave_list_connections` — enumerate routing primitives.
- `hookwave_create_destination` / `hookwave_create_connection` —
  provision new outbound targets and wire them to a source.
- `hookwave_list_events` / `hookwave_get_event` — read recent events
  with status, latency, headers, body.
- `hookwave_list_issues` / `hookwave_get_issue` — read auto-grouped
  delivery problems.
- `hookwave_replay_event` — re-deliver a past event without
  re-triggering the upstream provider.
- `hookwave_doctor` — pattern-matched failure diagnosis (signing
  mismatch, timeout, payload size, DNS / TLS, rate limits, persistent
  5xx). Returns ranked likely causes with concrete remediation.

## Optional

- [Model Context Protocol](https://modelcontextprotocol.io): the
  underlying protocol that connects AI clients to Hookwave's tool
  surface.
- [Anthropic tool-use docs](https://docs.anthropic.com/en/docs/build-with-claude/tool-use):
  what an LLM sees when it calls a Hookwave MCP tool.