# watcher.fyi

> watcher.fyi watches pages, JSON APIs, feeds (RSS, Atom, YouTube) and domains on a schedule (Free weekly, Pro daily, Max hourly). It diffs every change with deterministic probes (no LLM in the pipeline) and delivers each change as an FYI to a web inbox, signed webhooks, email, or your agent through a remote MCP server (`https://watcher.fyi/mcp`, OAuth 2.1 sign-in in the browser, no API key needed).

**If you are an agent:**
1. Connect over MCP OAuth. Add `https://watcher.fyi/mcp` and let the user sign in in their browser. Never ask the user for an API key.
2. Follow the [setup prompt](https://watcher.fyi/agent-start.md) (per-client commands for OpenCode, Claude Code, Codex, Cursor, VS Code, Claude and ChatGPT) to connect, verify with `get_plan`, and create the first watcher with the user.
3. Read the [agent playbook](https://watcher.fyi/docs/agents.md) for how to act on the user's behalf: inspect first, confirm before creating, never ack FYIs without consent, destructive tools only with the user's explicit OK and the `confirm` value, and what to say for each error.
4. Use the [recipes](https://watcher.fyi/docs/recipes.md) for exact `create_watcher` inputs, and the [MCP reference](https://watcher.fyi/docs/mcp.md) for every tool.

Anything the web app does, the MCP tools do too (account, billing links, channels, delivery and muting, inbox, API keys — list and revoke, never create — and connected agents). Over MCP these docs are also the resources `docs://{page}` and the `read_docs` tool.

Every public page is also available as Markdown: request it with `Accept: text/markdown`, or append `.md` (e.g. https://watcher.fyi/quickstart.md, https://watcher.fyi/index.md). The docs below are Markdown already. Everything in one file: [llms-full.txt](https://watcher.fyi/llms-full.txt).

## Start here

- [Agent setup prompt](https://watcher.fyi/agent-start.md): paste into your coding agent. It connects the MCP server over OAuth, verifies the plan, and sets up the first watcher with you.
- [Agent playbook](https://watcher.fyi/docs/agents.md): turning a vague request into a watcher, picking probes, reporting changes with `whats_new`, consent rules, and what to tell the user for each error code.
- [How watcher.fyi works](https://watcher.fyi/docs/guide.md): connecting agents (MCP OAuth per client, Settings → Agents, connected agents and Disconnect), targets and auto-detection, watchers, schedules and states, checks, retries and error streaks, probes, Change / Hold / Advance, item lists, request headers, FYIs, snapshots, retention, the complete plan limits table, downgrades, and the crawler.
- [Recipes](https://watcher.fyi/docs/recipes.md): goal → exact `create_watcher` input. Covers pricing pages, changelogs, GitHub releases, JSON API lists, job boards, tech stacks, status pages, docs sections, domain intel, sitemaps, YouTube, bot walls and low-noise text, each with the plan it needs.

## Reference

- [MCP reference](https://watcher.fyi/docs/mcp.md): server URL, OAuth 2.1 sign-in and scopes, every tool with inputs, outputs and annotations (destructive tools take `confirm`), resources (incl. `docs://`), prompts, parity with the web app, and structured error codes.
- [Probe catalog](https://watcher.fyi/docs/probes.md): the 18 probes by target kind, their plan tier, what each observes, config keys, and item-list options.
- [Delivery](https://watcher.fyi/docs/delivery.md): inbox, webhook and email channels, delivery rules (instant, daily or weekly digest; account defaults vs per-watcher; muting a watcher), digest schedule, a full webhook payload, and HMAC signature verification in JavaScript, Python and Go.
- [REST API](https://watcher.fyi/docs/api.md): base URL, auth (session and OAuth bearer tokens; API keys under Advanced), scopes, pagination, errors with `details` and `hint`, rate limits, and endpoints with curl examples.
- [OpenAPI spec](https://watcher.fyi/api/v1/openapi.json): the machine-readable REST schema, generated from the server.

## Policies

- [Bot](https://watcher.fyi/bot.md): how the crawler identifies itself, what it fetches, and how to block it.
- [Privacy](https://watcher.fyi/privacy.md): what is stored, for how long, and who processes it.
- [Terms](https://watcher.fyi/terms.md): acceptable use (only watch what you may access; no abuse or scraping at scale), billing, retention, liability, Australian law.

## Optional

- [Home](https://watcher.fyi/index.md): the landing page: probe list, delivery channels, plans and prices.
- [Human quick start](https://watcher.fyi/quickstart.md): four steps: sign in, add a watcher, get the FYI, connect your agent.
- [llms-full.txt](https://watcher.fyi/llms-full.txt): this index followed by the full text of every document above, in one file.
- [OAuth protected-resource metadata](https://watcher.fyi/.well-known/oauth-protected-resource): the RFC 9728 document for `/mcp`. The authorization server is the same origin.