Monitor Websites from Claude and ChatGPT with the AyeWatch MCP
AyeWatch's MCP server lets an assistant list, create, pause and delete monitors. Connect Claude, Claude Code or Cursor with an API key, and see where ChatGPT's OAuth route fits.
By AyeWatch Team5 min read
Short answer: AyeWatch's MCP server lets Claude and other assistants list, create, pause and delete monitors and manage your webhook. Connect it at https://ayewatch.ai/api/mcp with an API key in the Authorization header. ChatGPT needs AyeWatch's OAuth sign-in instead, which is limited to registered apps. A paid plan is required.
What the AyeWatch MCP server lets an assistant do
The endpoint is https://ayewatch.ai/api/mcp. It uses Streamable HTTP, needs no session ID and is stateless. It manages the same monitors as the website, the mobile app and the REST API, with the same plan limits and credits. The table below lists its eight tools.
You can say what to monitor in a sentence, and the assistant turns it into a monitor. Updates still reach you the usual way, through your notifications and webhook. The tools manage monitors and the webhook, and the docs state they do not provide an alert-history feed. Ask the assistant to set something up, not to summarize last week's alerts.
Connect with an API key
Create a key in API access settings. The docs say API and MCP access come with every paid plan. Send the key as Authorization: Bearer aw_live_... on every request.
Claude Code takes the header on the command line:
claude mcp add --transport http ayewatch https://ayewatch.ai/api/mcp \
--header "Authorization: Bearer aw_live_YOUR_API_KEY"
Cursor reads this entry from its MCP configuration, which is the example in AyeWatch's docs:
{
"mcpServers": {
"ayewatch": {
"url": "https://ayewatch.ai/api/mcp",
"headers": { "Authorization": "Bearer aw_live_YOUR_API_KEY" }
}
}
}
Claude (the apps) can add a custom connector. Anthropic's help pages say that for a server that uses an API key you choose No sign-in and add the key under Request headers, with the header name Authorization and the value Bearer aw_live_YOUR_API_KEY. Anthropic describes request headers as a beta that only some organizations have, so the option may be missing from your dialog.
Keep the key private. It reaches your monitors and your webhook settings, so use a separate key for each client and revoke one if it leaks.
Where ChatGPT fits
OpenAI's help page for custom MCP servers lists OAuth and No authentication as the sign-in choices, and it has no field for an API key header. That rules out the key method above. AyeWatch's own OAuth sign-in is limited to registered apps, and its docs add that this does not mean AyeWatch is listed in an app directory. So the route into ChatGPT is a connection that offers the AyeWatch sign-in. If you do not see one in your workspace, use Claude, Claude Code or Cursor with a key instead. ChatGPT's menus for this change often, so check OpenAI's help page for the current steps.
On the AyeWatch side, OAuth shows a consent screen. Read access lets the app view your monitors, and write access lets it create, change, pause and delete them. Access expires after 30 days or earlier if you disconnect, and Connected apps in settings lets you disconnect. OAuth connections cannot touch the webhook, because the webhook receives every alert and its secret signs them.
A first request that is safe to run
Sample rule, as you would type it: Create a monitor for https://northgatedental.example/pricing. Alert me when the price of a cleaning, whitening or filling changes. Create it paused, show me what you made and wait for my go-ahead before you turn it on.
The assistant should call create_topic with arguments like these, which follow the example in the docs:
{
"name": "https://northgatedental.example/pricing",
"description": "Alert me when the price of a cleaning, whitening or filling changes.",
"monitoring_mode": "auto",
"is_active": false
}
A paused monitor does not run checks, and the docs say active monitors consume credits as checks run. After you agree, the assistant calls update_topic with the returned ID and changes: {"is_active": true}. When the page changes, the alert reaches you like this: "Northgate Dental raises whitening to $299. The pricing page lists teeth whitening at $299, up from $249. Cleaning and fillings are unchanged."
When MCP is the wrong tool
- Scripts and schedules. A cron job should call the REST API directly. MCP is for an assistant that decides what to call.
- Reading alert history. No tool returns past alerts. Use the app feed, or a webhook receiver.
- Bulk edits you cannot undo.
delete_topicpermanently deletes a monitor and its schedule. Ask the assistant to list what it will delete before it acts. - Heavy traffic. MCP and REST share the key's limit of 120 requests per minute by default, and every MCP POST counts, including initialization and tool discovery. Batches are rejected.
- Pages behind a login. AyeWatch reads public pages and cannot sign in.
Cost and effort
AyeWatch Pro is $9/mo with 50 topics and hourly checks; Pro+ is $19/mo with 200 topics and ASAP mode. Every plan starts with a 7-day free trial. Most checks use 1 credit. The monitors an assistant creates use the same credits as monitors made in the app, and the docs list no separate bill for API access. Setup is one command or one config entry. Claude and ChatGPT have their own plans, so check their pricing pages for what a connector needs.
Connect an assistant in five steps
- Create an API key in API access settings and copy it once. Use one key for each client.
- Add the server to your client: the claude mcp add command for Claude Code, the JSON entry for Cursor, or a custom connector with Request headers in Claude.
- Ask the assistant to list your monitors. This uses list_topics and confirms the key works.
- Ask it to create one monitor, paused, using the sample rule above, and read what it created.
- Tell it to turn the monitor on, then confirm the first alert arrives through your normal notifications.
| Tool | What it does | Worth knowing |
|---|---|---|
| list_topics | Lists monitors | Filters by page, page_size, is_active, interval and dates. Default page size 20, maximum 100. |
| get_topic | Gets one monitor | Takes a numeric topic_id |
| create_topic | Creates a monitor | Takes a name, page URLs or a platform source, plus an interval or monitoring_mode auto |
| update_topic | Changes a monitor | Send changes with only the fields to update. is_active false pauses, true resumes. |
| delete_topic | Deletes a monitor | Permanent, and it removes the schedule |
| get_webhook | Reads the webhook URL and whether delivery is on | API key connections only |
| set_webhook | Creates or changes the webhook | API key only. AyeWatch emails the account owner when the URL changes. |
| rotate_webhook_secret | Replaces the signing secret | API key only. The old secret stops working at once, and the new one is never returned. |
Start a monitor in under a minute
Describe a topic or paste a URL. Cancel before the trial ends and you pay nothing.
Try it free for 7 daysThe bottom line
Create a key, add the server to Claude Code or Cursor, and ask for one paused monitor on a page you check by hand. Read what it created before you tell it to start.
FAQ
What is the AyeWatch MCP server URL?
https://ayewatch.ai/api/mcp. It uses Streamable HTTP, and it is stateless, so no session ID is needed.
Which AyeWatch plans include MCP access?
Every paid plan. API keys are only available on paid plans, and the docs say API and MCP access come with each of them.
Can Claude read my past alerts through MCP?
No. The tools manage monitors and the webhook and provide no alert-history feed. Alerts still arrive through your notifications and webhook.
Does OAuth share my API key with the app?
No. You sign in to AyeWatch and approve read or write access, and the app never sees a key. Access expires after 30 days or sooner if you disconnect.
How many MCP requests can I send?
120 per minute per API key by default, shared with the REST API. Each MCP POST counts once, including initialization and tool discovery.