Today, we're introducing web search monitors.
A page monitor watches one URL. A web search monitor watches the whole web. Give it a search query and a plain-language description of what you care about, pick a schedule, and Context runs the search for you on every check. When a result shows up that you haven't seen before and it describes a real new development, you get an alert.
Most of what's worth knowing doesn't happen on a page you already know to watch. A new model drops on a lab's blog. A competitor announces a feature in a press release. A regulator publishes a rule on a site you've never visited. Web search monitors catch those, so you (and your agents) hear about them first.
What people are watching
Here are some of the monitors we've seen people set up, and the ones we're running ourselves:
- New AI model releases. "Alert me when a major lab releases a new model or opens API access to one." Find out the day it ships, not the day it trends.
- The latest on a breaking story. Follow an ongoing news story and get each genuine update, without the recaps and opinion pieces that repeat what you already know.
- Competitor launches and pricing changes. Watch for new products, features, and pricing announcements from the companies you compete with.
- Funding rounds and acquisitions. Track deals in your market, or know the moment a prospect raises a round and has budget to spend.
- Mentions of your company. Catch reviews, press coverage, and forum threads about you as they're published.
- Research papers on a topic. Get new papers on retrieval, agents, or whatever you're working on, as they're posted.
- Security advisories. Hear about new vulnerabilities and CVEs affecting the libraries and vendors you depend on.
- Regulation and policy. Follow new rules, guidance, and enforcement actions in your industry or region.
- Hiring signals. Spot companies opening new offices or hiring for roles that suggest they need what you sell.
- Product drops and restocks. Know when a product you're waiting for launches or comes back in stock.
If you can describe it in a sentence, you can monitor it.
Only what's actually new
A naive "search and diff" alert gets noisy fast. The same story gets rewritten by fifty outlets, evergreen pages drift into the results, and old articles get re-dated.
Web search monitors use semantic detection by default. Each result you haven't seen before is judged against your instructions, and evergreen pages, stale content, and re-reports of something the monitor already alerted you about are filtered out. Want every new result instead? Switch to exact detection and you'll get them all.
Queries support operators like site:, -site:, and quoted phrases, so you can focus a monitor on specific sources or exclude the ones you don't trust.
Each change lists the new results in matched_urls with the evidence behind the alert, and every run is kept in the monitor's history, so the feed works just as well as an event source for agents and data pipelines as it does for notifications.
Set them up through the MCP
The fastest way to create a web search monitor is to ask your agent. With the Context MCP server installed in Claude, ChatGPT, Cursor, or any other MCP client, just say what you want to know about:
Set up a monitor that checks every hour for new AI model releases from the major labs, and post alerts to my Slack.
Your agent creates the monitor, picks the schedule, and wires up the webhook. Slack webhook URLs get nicely formatted messages out of the box. You can list, edit, pause, and run monitors through the MCP the same way.
Install the Context MCP server to get started.
Or use the API
Here's a monitor that watches for new model releases every hour. Store your API key in CONTEXT_DEV_API_KEY and run:
curl https://api.context.dev/v1/monitors \
--request POST \
--header "Authorization: Bearer $CONTEXT_DEV_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"name": "New AI model releases",
"target": {
"type": "search",
"query": "new AI model release",
"instructions": "Alert me when a major AI lab releases a new model or opens API access to one. Ignore rumors, benchmarks of existing models, and opinion pieces."
},
"schedule": { "type": "interval", "frequency": 1, "unit": "hours" },
"webhook": { "url": "https://example.com/webhooks/context" }
}'The first run captures a baseline right away. From then on, you get a change.detected event whenever something new matches.
Pricing
Each run costs 2 credits, the same as a 20-result web search. Creating and managing monitors is free, and every plan includes monitors, including the free plan. Schedules run as often as every 10 minutes.
Start monitoring
Pick the three things you'd most hate to hear about late, and set up a monitor for each one. It takes about a minute.
Sign up for Context.dev and create your first web search monitor today. Then install the MCP server and let your agent set up the rest.