t3-watch MCP server

Read from the package source without running it. The installed server may list more.

• watch_alerts

  Where change events go, if anywhere.

• watch_apply

  Applies the last scan: turns on the listed checks (or every suggested one) and removes the listed keys by access key.

• watch_check

  One check's definition, recent runs with evidence, notes and changes. Use it to investigate.

• watch_decide_draft

  Approves or discards a draft. Only the Watch view calls this, on the person's click.

• watch_draft_check

  Drafts a check. Watch validates it and runs it once; it stays a draft until the person turns it on. Kinds and config: - http: { url, expectStatus?, contains? } - github: { repo: "owner/name", branch?, workflow?, maxAgeSeconds? } (maxAgeSeconds needs workflow and also fails when that workflow's newest successful run is older, for scheduled workflows) - aws_metric: { profile?, region, namespace, met

• watch_draft_vendor

  Drafts a vendor description (JSON) so Watch can recognise and check a new API-key vendor. Validated, tested with a stored key when there is one, and kept as a draft until the person approves it.

• watch_note

  Leaves a short note on a check (or on Watch as a whole), shown on Home next to it. Use it for what you found.

• watch_runs

  Recent runs of one check, for the Watch view.

• watch_scanwrite action

  Runs setup: reads the repositories under the folders and the logins the person allowed on the page (see watch_sources), stores working keys, proposes checks and runs each once. Refuses until the person has chosen. Takes about half a minute.

• watch_set_alerts

  Sets where Watch sends a change event (one POST per check that starts or stops failing): normally a T3 webhook task URL. Pass url null to stop. Watch signs each body in X-Watch-Signature with the secret the person made on the page; this tool never takes a secret.

• watch_set_check

  Turns a check on, pauses it, snoozes it until a time, resumes it, or runs it now.

• watch_set_product

  Names the product a service belongs to on Home, for when Watch's guess from names is wrong, such as a site whose repository has another name. A site or repository with no product of its own joins a named product whose name matches its own, and a repository follows the site it matches. Names compare without case, accents or punctuation; a product has one spelling, so naming it in a new one respells

• watch_sign_in_link

  Makes a link that signs a browser in to the Watch page. It works once, within … minutes. Make one only when the person asks to open the page, and give it only to them.

• watch_sources

  What setup may read: the folders, each with how many repositories it holds, and the logins (the GitHub CLI, AWS CLI profiles, keys in .env files), each with whether it is on this machine and allowed. The person changes this on the page, before a scan.

• watch_start_scan

  Starts setup in the background, for the Watch view.

• watch_state

  The full state for the Watch view.

• watch_status

  What is broken, what could not run, drafts waiting, and how many checks pass.

• watch_test_alertwrite action

  Sends one test alert, signed when a secret is set, to the webhook Watch's alerts go to, and says whether the webhook accepted it. It wakes the thread that webhook is bound to, once. Use it right after watch_set_alerts, and when the person asks whether alerts work.

• Code scan63 source files scanned25/25
• –Live reliabilityno gateway calls yet and no remote to proben/a
• –Tool poisoningtools not inspected (local package is not executed); not countedn/a
• Auth qualitylocal package, no credentials required12/15
• Maintenancelast push 0 days ago15/15
• Maintainer identitynamespace and repository owner differ; GitHub account older than a year5/10

Safety score

88/100

higher than 80% of 4,460 AI and agents servers

Tools

18

the typical AI and agents server has 6

Tools that change things

11%

typical: 0%. Fewer write tools means less an agent can break

Downloads a week

577

more than 88% of 1,538 AI and agents servers with a package

t3-watch: common questions

Is t3-watch MCP server safe?

Yes, by our scan: it is graded A (88/100). Read the t3-watch safety report

How do I install t3-watch?

It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.

Does t3-watch need an API key?

Not as far as the registry entry and our scan can tell: no credentials are declared or required.

Is t3-watch maintained?

The last commit was 11 days ago (2026-10-24). The latest release is v0.6.0.

What can I use instead of t3-watch?

Servers from other publishers that do the same job: Claude Real Video MCP server, Power BI Authoring MCP Server (Local) and Watch Skill MCP server.

How does t3-watch compare with other AI and agents MCP servers?

safety score 88/100 (higher than 80% of 4,460 AI and agents servers); tools 18 (the typical AI and agents server has 6); tools that change things 11% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.