MCP server
A process exposing tools, resources or prompts over the Model Context Protocol. One well-built server — for a database, an API, a filesystem — becomes usable by every MCP-capable client without further glue.
One server, every compliant client
A server is a process exposing capabilities over the protocol — a database, a filesystem, a SaaS API, an internal service. Build it once and every MCP-capable client can use it, which is a materially different economic proposition from writing an integration per product.
What a good one exposes
Tools with narrow, well-named operations and honest schemas; resources the client can read; and errors that say what went wrong rather than returning an empty result. The most common failure is a server that exposes one god-tool with twenty optional parameters, which models use badly and humans cannot audit.
Permissions belong at the boundary
A server is the right place to enforce what may be read and written, because it is the only component that understands the underlying system. Relying on the model to restrain itself is not a security model; scoping the server’s own credentials is.
In AGNT
Connect MCP servers to AGNT and their tools appear in agent toolkits like any native node.
MCP server — common questions
Do I need to write an MCP server to use AGNT?
No. AGNT ships a large library of built-in tools and integrations; MCP servers are an additional source when you have a system that needs one.
Can an MCP server run locally?
Yes, and many do — a local server keeps the data path entirely on your machine, which pairs naturally with a local-first runtime.
How do agents know what a server can do?
The server advertises its tools and schemas on connection, and the runtime presents them to the agent as callable capabilities.