ForgingStack Watchtower
Integrates with ForgingStack Watchtower to create and manage agent liveness watches, allow agents to check in, retrieve watch status and credit balances, and fetch hash-chained alert or recovery receipts. Supports overdue alerting when agents miss check-in intervals and snoozing watches to silence alerts.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ForgingStack Watchtower Create a watch for my deploy agent, check in every 5 min, alert ops@example.com"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ForgingStack Watchtower
The agent that stops checking in IS the one you will hear about.
mcp-name: io.github.jdkabattles/forgingstack-watchtower
This is the public client for Watchtower. The service itself — the liveness checks, the alert emails, the receipt chain, the key store, and billing — runs on ForgingStack's private service and is not in this repo. Copying this repo does not give you the service: you still need an API key and credits, from signup below.
Live API: https://forgingstack-watchtower.jkbattles.workers.dev
How it works
You register a watch on an agent: how often it should check in (
interval_seconds, 30 to 2,592,000) plus extra quiet time allowed (grace_seconds, 0 to 86,400), and where the alert email goes. One watch per agent per owner. Registering a watch costs 1 credit and returns a check-in key, shown once.The agent checks in with its watch id and check-in key. Check-ins are free and need no API key.
If the agent goes quiet past its interval plus grace, Watchtower emails an alert and issues a hash-chained receipt. Each alert costs the watch owner 1 credit; an owner at 0 credits is skipped and the watch stays overdue.
When the agent checks in again after an alert or overdue period, a recovery receipt is issued. You can also snooze a watch (1 to 1,440 minutes) to silence alerts for a while.
Related MCP server: Tiltmeter
Signup (v1)
curl -X POST https://forgingstack-watchtower.jkbattles.workers.dev/v1/signup \
-H "Content-Type: application/json" \
-d '{"email":"you@example.com"}'Returns an API key once and 100 starter credits. Save the key immediately — it is stored only as a hash and cannot be shown again. Click the verification link in the signup email before the key can register watches; until then, watch registration returns 403.
Use it (MCP, stdio)
Set the key, then run the MCP server in this repo:
export FORGINGSTACK_API_KEY="your-key-here"
python -m app.mcp_serverTools: create_watch, check_in, get_watch, get_credits,
get_receipt. See SKILL.md for when to use it and when not to.
Price
1 credit to register a watch, 1 credit per alert, check-ins free.
100 starter credits on signup. Credit packs: $10 for 1,000 credits,
bought with POST /v1/billing/checkout — credits are added once the
payment is verified (usually within the hour).
Files in this repo
app/mcp_server.py— MCP wrapper client (stdio, stdlib only)SKILL.md— when to use it, how to read watch statusagent-card.json— A2A-style discovery cardserver.json— MCP Registry metadatapyproject.toml— package metadata for the MCP client
Available Tools
5 toolscheck_inB
Check in for a watch. Free. Needs only watch_id and the check-in key from watch registration — no API key. A check-in after an alert or overdue period returns a recovery receipt.
Inputs: watch_id (required), checkin_key (required).
create_watchA
Register a liveness watch on an agent. The agent must check in at least every interval_seconds (plus grace); if it goes quiet past that, Watchtower emails an alert and issues a hash-chained receipt. Costs 1 credit. Returns the watch plus a check-in key, shown once — save it: the agent checks in with it, and it cannot be shown again.
Inputs: agent_id (required), notify_email (required), grace_seconds (required), interval_seconds (required).
get_creditsB
Remaining credit balance for the configured API key.
Inputs: no named parameters.
get_receiptB
Fetch a past Watchtower receipt by id. Receipts are hash-chained: each receipt's hash covers the previous receipt's hash, so tampering breaks the chain.
Inputs: receipt_id (required).
get_watchB
Fetch a watch by id, with its current status (healthy, overdue, alerted, snoozed) and its last 20 events.
Inputs: watch_id (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool targets a distinct resource and action: create/check-in/get watch, get credits, and get receipt. There is little risk of confusing them, even though get_watch and get_receipt both fetch by id.
Names are consistently snake_case and mostly follow a verb_noun pattern. The only minor deviation is check_in, which is a phrasal verb rather than a strict verb_noun form.
Five tools is well-scoped for a focused liveness-watch service. Each tool maps to a clear operational need: registration, check-in, inspection, credits, and receipts.
The core watch lifecycle is covered, but notable operations are missing: no list_watches, cancel/delete watch, update interval, or snooze action despite status including snoozed. These gaps may force agents to work around the surface.
Maintenance
Related MCP Connectors
Manage uptime monitors, incidents and status pages from any MCP client.
Hosted MCP memory and agent control plane for durable conversations, jobs, and operations.
Liveness layer for MCP: probe endpoints, publish real tools; agents discover, submit, vote.
Workflow diagnostics, capability routing, and x402 settlement for MCP-compatible agents.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables MCP clients such as Claude to check current uptime status, inspect monitors and incidents, trigger fresh checks, and retrieve status summaries from a self-hosted monitoring service.2MIT
- AlicenseNot gradedqualityBmaintenanceEnables MCP clients to check whether third-party API responses have drifted, inspect expected schemas, review change history, list monitored endpoints, and trigger checks before agents trust an API.278 PyPI1MIT
- AlicenseNot gradedqualityCmaintenanceEnables MCP clients to check a resident's wellness score, hours since last activity, and alerts; retrieve recent front-door events; and send simulated family check-in notifications.6 npmMIT
- AlicenseNot gradedqualityBmaintenanceEnables MCP clients and webhook-driven agents to exchange durable messages with atomic claim, completion, and acknowledgement workflows, plus verified webhook event subscriptions and native wake-ups.MIT