MCP Servers Explained

An MCP (Model Context Protocol) server gives the AI access to external tools — a standard way for an AI model to call a tool or data source outside Kamo, instead of every integration needing its own wiring.

Two transport types

• Streamable HTTP — the current standard. The AI's requests go to one HTTP endpoint on the server. Recommended.
• Server-Sent Events (SSE) — the older standard, for servers that have not moved to Streamable HTTP yet.

Programs that run on a computer and talk over standard input/output (stdio) are not supported: Kamo never runs another program on its servers. Host such a server behind Streamable HTTP and add its URL.

Tool names

Every tool from a server gets the server's prefix, such as ext_docs_, so two servers can offer a tool with the same name. The prefix never changes once the server is added.

Who can use the tools

Only a server's read-only tools reach anyone. A tool is read-only when its server marks it so; tools that change something are offered to no one, neither to people nor to AI members.

• People in your organization can use the read-only tools of every active server with the AI Assistant, unless their access policy turns off Allow MCP Tools (see AI Security Models Explained).
• AI team members can use a server's read-only tools too, but only when Available to AI members is on.

Your Kamo sign-in is never sent to an MCP server: it receives only the headers you configure.

Health and tool count

Kamo checks each server every ten minutes and when you test it. A healthy server shows how many tools it offers; an unhealthy one shows the last error.