Every server, with the tools and data it exposes.
Read the risk, owner, status and connected data sources for each MCP server in your register.
Scroll the image to inspect the details.
From the server to the policy.
A changed tool description, caught.
The description of mcp-database-connector’s execute_query tool was changed to tell agents to send the connection string to an outside address. Guard0 raised it as critical and contained it.
Scroll the image to inspect the details.
Our census reached 2,958 MCP servers on the internet. 1,139 of them answered anyone, with no authentication.
From the config file to the tool call.
- Every server, local and remote
- MCP servers in laptop configurations, platform connectors and code, each with the client that uses it.
- Tools and credentials
- The tools each server exposes and the credential it holds: a static key, OAuth, or none at all.
- Checks before release
- MCP declarations in your repositories are checked with each pull request, and critical findings go to the agent’s owner.
- Guards on tool calls
- Allow, redact, coach or deny a call where you turn guards on, with every decision on the record.
Common questions
Does Guard0 find remote MCP servers as well as local ones?
Yes. Local servers come from laptop configurations through the endpoint sensor. Remote servers come from the clients, platforms and code that connect to them.
Can Guard0 block a tool call?
Yes, where you turn guards on. A guard can allow, redact, coach or deny a call, and every decision goes on the record.
Start with one team’s MCP servers.
Roll out the sensor to one team, or connect one repository, and see every server your agents connect to.