MCP Server

Connect AI agents to HookWatch via the Model Context Protocol. Manage webhook endpoints, retry failed events, create cron jobs, and configure alerts — all from your AI coding assistant.

How It Works

The MCP server acts as a bridge between AI agents and the HookWatch platform. It connects to the HookWatch core via gRPC and exposes 34 tools that agents can call.

AI Agent (Claude Code, Cursor, Windsurf, etc.)
    |  MCP Protocol (stdio or HTTP)
    v
HookWatch MCP Server
    |  gRPC (authenticated with your API key)
    v
HookWatch Core API

Two transport modes are supported:

  • stdio — runs locally as a subprocess of your AI tool (recommended for development)
  • HTTP — hosted server at mcp.hookwatch.dev (no local install needed)

Setup

Step 1: Create an API Key

Go to the HookWatch DashboardAPI KeysCreate Key. Give it a name like mcp-server and copy the key.

Store your key securely

The full API key is only shown once. If you lose it, create a new one.

Step 2: Connect Your AI Tool

Claude Code

Option A: Local binary (stdio)

Terminal

Option B: Hosted server (HTTP — no install needed)

Terminal

Cursor

Option A: Local binary (stdio)

.cursor/mcp.json

Option B: Hosted server (HTTP)

.cursor/mcp.json

Other MCP Clients

Any MCP-compatible client can connect via HTTP. Use the endpoint https://mcp.hookwatch.dev/mcp with an Authorization: Bearer hk_... header.

Install Binary (Optional)

Only needed for stdio mode. If you use the hosted HTTP server, skip this step.

Terminal

Also available for macOS (darwin_arm64, darwin_amd64) and Windows (windows_amd64.zip).

Self-Host with Docker

Run your own MCP server instance. Clients authenticate with their own API keys via the Authorization header — no server-wide key needed.

Terminal

The server exposes /mcp (MCP endpoint) and /health (health check) on port 8085.

Available Tools

34 tools across 7 categories:

Endpoints

ToolDescription
list_endpointsList all webhook endpoints with stats
get_endpointGet endpoint details, retry and signature config
create_endpointCreate a new webhook endpoint
update_endpointUpdate name, URL, active status, or retry config
delete_endpointDelete an endpoint and its events

Events

ToolDescription
list_eventsList events with status, endpoint, and pagination filters
get_eventGet full event details (headers, body, response)
get_event_metricsAggregated delivery metrics with daily breakdown
count_eventsCount events by endpoint or status
retry_eventRe-deliver a failed event
replay_eventReplay with optional modified body/headers
delete_eventDelete an event

Alerts

ToolDescription
list_alertsList all alert configurations
get_alertGet alert details
get_alerts_by_resourceFind all alerts for an endpoint or cron job
create_alertCreate alert (Slack, Discord, email, Telegram)
update_alertUpdate alert target, type, or template
delete_alertDelete an alert

Cron Jobs

ToolDescription
list_cron_jobsList all cron jobs
get_cron_jobGet cron job details
create_cron_jobCreate a new cron job
update_cron_jobUpdate schedule, command, or config
delete_cron_jobDelete a cron job
enable_cron_jobEnable a disabled job
disable_cron_jobDisable a job
run_cron_jobTrigger manual execution
list_cron_executionsExecution history with filters
get_cron_executionExecution details (stdout, stderr, exit code)
retry_cron_executionRetry a failed execution
list_agentsList registered CLI agents

Teams, WebSocket & API Keys

ToolDescription
list_teamsList your teams
get_teamGet team details
list_team_membersList team members
list_ws_endpointsList WebSocket proxy endpoints
get_ws_endpointGet WebSocket endpoint details
list_ws_connectionsList active WebSocket connections
list_api_keysList API keys (values hidden)
create_api_keyCreate a new API key
delete_api_keyDelete an API key

Examples

Once connected, your AI agent can call any tool. Here are some common workflows:

List your endpoints

MCP

Create an endpoint

MCP

Retry a failed event

MCP

Open Source

The MCP server is open source at github.com/hookwatch-dev/mcp-server. Contributions are welcome.

Related