Founding-customer offer: Sign up before January 1, 2027 and get 20% off your first year of MCP Tower. Lock in 20% →
20% off your first year. Join before 2027. Claim →
Change one URL and MCP Tower shows which AI agents use your MCP server, which tools they call, and what users do next.
Companies ship an MCP server to get listed in agent directories, and then nobody can say whether it gets used. Finding out turns into a project for four teams, and most never finish it.
MCP Tower replaces that project with a URL change: no SDK, no code change on the MCP server, no data pipeline to own.
MCP Tower is a hosted proxy in front of your remote MCP server. Every request passes through to your server unchanged, and MCP Tower records what happened.
Add your MCP server's URL in MCP Tower. Any remote MCP server reachable over HTTPS works, whatever it is built with.
Give agents and catalogs the MCP Tower URL instead of your server's. If you are already listed, resubmit with the new URL.
See which agents call which tools, how fast they answer and where they fail, as the calls come in.
MCP Tower goes live before all of its parts do. Each part below is marked until it ships.
Which agents call which tools, how often, how fast and with what errors, plus a live view of calls as they happen.
Widgets that agents render report clicks and form submissions back to your dashboard.
Links in tool responses are tracked, so you learn which ones users open and which agent sent them.
Checks that understand MCP confirm your server answers and every tool responds, even when no agent is calling.
Answer the catalog questions once, reuse them everywhere, and get the manifest file for OpenAI's upload flow.
A catalog shows installs, but you can't tell whether anyone actually uses your server, or a client asks what they got for the project.
Usage and uptime for every server in one place, with nothing to build or maintain.
Someone asks how the MCP integration is doing, and the only answer is that it's live.
Numbers you can put on a slide and an alert before a customer reports a problem.
AI agents are a new channel, and your web analytics can't see any of it.
AI agents as a measurable traffic source, and a faster way into more catalogs.
You were asked to add analytics, or a customer found an outage before your monitoring did.
Debugging and monitoring without building a pipeline or adding an SDK.
Register your server, publish one URL, and watch the calls come in. Questions first? Write to us.
Yes. Register your server in MCP Tower and publish the MCP Tower URL instead of your own. If your server is already listed in a catalog, resubmit the listing with the new URL. Nothing changes on your MCP server.
It works with any remote MCP server reachable over HTTPS, whatever framework or language it is built with. No SDK is required. Local (stdio) servers are not supported, because their traffic never crosses the network.
The proxy runs on Cloudflare's global network. Recording an event is fire-and-forget: it never blocks or breaks the request it records, so a problem with analytics never affects your traffic. Health checks watch both MCP Tower and your server. A self-hosted nginx module that takes the hosted proxy out of the path is on the roadmap.
Yes. MCP Tower supports the Streamable HTTP transport, with both JSON and SSE responses. The deprecated HTTP+SSE transport is not supported.
Metadata for each call (tool name, agent, timing, status and error class) and the identifiers needed to attribute it: the AI agent, a user id and a session id. Tracking is opt-in per tool, so a tool you have not enabled, including one your server adds later, is not tracked. You choose which tool arguments are tracked; nothing else from the arguments is stored.
The proxy forwards authorization headers and does not store them. Agents' tokens are never stored. To scan a server that needs OAuth, MCP Tower signs in as its own OAuth client with your consent; those tokens are stored per user, encrypted, and you can reset them at any time.
No, as long as your live server matches what you submitted: the URL, the list of tools and their descriptions. That is why the submission kit includes the MCP Tower URL and the tracking tool, and why MCP Tower passes tool descriptions through unchanged. MCP Tower also warns you when your live tools drift from what you submitted.
From the caller's network: MCP Tower maintains the ASNs and IP ranges each AI agent calls from and keeps that list current as agents change their infrastructure, so you don't have to.
Yes. MCP Tower rescans every server daily. When a tool or an argument is added, changed or removed, down to a changed description, Admins and Editors get an email and the change stays highlighted until someone reviews it.
You can log tool calls yourself. What logging can't easily show you is what happens around them: widget interactions, link clicks after the answer, a breakdown by agent, uptime measured from outside your network, and catalog guidance. And you would still have to build the pipeline and the dashboards. MCP Tower gives you the whole chain in an afternoon.