Uptime Monitoring From Your AI Agent.
Connect Claude, ChatGPT, Cursor or Codex to FlareWarden's MCP server. Ask whether a site or a service like Stripe is down, then have your agent plan your sites, the services they depend on, cron jobs, status pages and reports. It proposes, you confirm, and it can only do what you can do on the website.
https://mcp.flarewarden.com/mcp
Streamable HTTP. Works with Claude, Claude Code, ChatGPT, Cursor, Codex and other MCP clients.
Add FlareWarden to Your Agent.
One server address for every client. Claude, Claude Code, ChatGPT, Cursor and Codex all sign you in on FlareWarden, so your agent acts as you; scripts, and any client you'd rather give a token, can use a personal token from the AI Agents & API page in your account.
In Claude's Connectors settings, add a custom connector and paste the server address:
https://mcp.flarewarden.com/mcpClaude sends you to FlareWarden to sign in and approve the connection the first time you ask about your account.
Personal tokens expire after 30 days, 90 days or a year, as you choose, and are shown once. Step-by-step instructions for each client are in the setup guide.
Ask Before You Sign Up.
Connect the server and these work right away, with no sign-in. Here is what your agent asks FlareWarden, and what comes back.
Stripe's own status page and FlareWarden's checks from several regions, side by side; the verdict, which is the worse of the two; open incidents; and 7, 30 and 90-day uptime, with a link to the service's page on flarewarden.com/status. It covers hundreds of services, by name or domain.
The address is requested once from each of 6 locations (Ashburn, San Jose, London, Frankfurt, Singapore and Sydney). Back comes each one's result (up, down, blocked or inconclusive), the status code and timings, and an overall verdict. A host is checked at most once every 5 minutes, and the page's content is never returned.
A preview of the site's monitoring: the third-party services FlareWarden recognizes from its DNS records and home page (hosting, CDN, DNS, payments, email, analytics), each with the evidence and its status now; when its SSL certificate expires; a summary of its DNS; whether it is up from 6 locations; and what FlareWarden would set up, in monitors, on each plan.
Your agent asks whether you have a FlareWarden account. If you don't, it gives you the sign-up link straight away; signing up carries nothing from the conversation. Come back once you've signed up: your agent signs you in, shows the plan and, once you say yes, sets it up. Want to see what it would cover first? Ask for the preview.
Your agent can also ask what FlareWarden does and costs, and search these docs, without an account. Public checks are explained on FlareWarden's public checks page.
Plan It in Chat. Confirm, Then It's Done.
The first time you ask about your own account, your agent asks you to sign in to FlareWarden and approve the connection. New monitors and status pages start as a plan with recommendations and their cost, and nothing is created until the plan's questions are answered.
A plan first: what the site runs on (hosting, CDN, cloud regions, payments), the page checks that would hold on it, and each choice with a recommendation and its cost in monitors. Up to 50 sites at once, for agencies.
How AWS will be watched, then the choices in order: which regions (recommended from your site's DNS), which components count, how serious each one is, and how often to check.
The schedule as understood, with the next five runs in your time zone and a recommended grace period. Once you confirm, the job's ping URL and code to paste for crontab, GitHub Actions, a Kubernetes CronJob or a systemd timer.
Uptime over the time FlareWarden measured, downtime minutes, incidents, the longest incident and mean time to recover, p50 and p95 response times, and certificate and domain expiry, for a site, a group or the whole account.
A draft: the exact text, every status page it will appear on, and how many subscribers, members and webhooks will be told. It is posted only when you approve it. Maintenance windows work the same way.
A plan with the page's address, its components and your plan's limits. A status page is public the moment it exists, so publishing it is always your decision, never the agent's.
It also covers content checks, DNS watches, pausing and tuning monitors, your team, alert emails and webhooks.
Your Agent Plays by Your Rules.
The agent goes through the same permission checks and plan limits as the website, on every call.
Never more than you
It acts as you: your role, your access to sites and status pages. What the website would refuse you, it refuses the agent.
Plans before changes
New monitors are shown to you as a plan with their cost, and created only once you confirm. Incident posts and maintenance are drafted, then published on your yes.
A daily budget
Each connection may add at most 25 monitors a day unless you change it, when you approve the connection or later.
Secrets stay out of the chat
An API token a dependency check needs is entered on a FlareWarden page, never in the chat. A new webhook's signing secret is shown once, on FlareWarden.
Deletes and billing stay with you
Deleting sites, checks or status pages, and anything to do with billing, is done in the dashboard. The agent points you there.
Off when you say so
Revoke a connection at any time. The account owner can turn off AI agent access for the whole account, which ends every connection. The audit log shows what was done through an agent.
On every plan, Free included.
Connecting an agent costs nothing extra. Monitors it adds use your plan's monitor pool, as they would from the dashboard, and your plan's limits apply as they do on the website. The Free plan has 15 monitors and needs no credit card.
Frequently Asked Questions
MCP (Model Context Protocol) is an open standard that lets AI assistants use outside tools. FlareWarden's MCP server, at https://mcp.flarewarden.com/mcp, gives Claude, ChatGPT, Cursor, Codex and other MCP clients tools to check whether a website or service is down, and to read and manage your FlareWarden monitoring.
It is in beta. It works on every plan, and its tools and their answers may still change as we improve them. Send feedback or report a problem through the contact form at flarewarden.com/contact or to support@flarewarden.com.
Every plan, Free included. Monitors your agent adds use your plan's monitor pool, as they do when you add them from the dashboard.
Not to ask whether a service or website is down, what to monitor on a site, or how FlareWarden works. Your own sites, reports and status pages need an account; your agent asks you to sign in the first time you ask about them.
No. Deleting sites, checks and status pages, and billing, stay in the dashboard. Within what your role allows, your agent can pause and resume monitors, change their settings and remove a component from a status page.
Only what that teammate can see on the website. An agent connects as a member, with that member's role and site access, checked again on every call, so removing someone from the account also ends their agent's access.
Yes. Connect to https://mcp.flarewarden.com/mcp/readonly for reading only, or grant only the Read permission when you approve the connection.
Ask Your Agent. Know What's Down.
Start with the free plan, connect your agent, and let it plan your first site with you.