MCP server

An MCP server is the side of the Model Context Protocol that offers capabilities to AI applications. An MCP client inside the application connects to it, lists what it offers, and lets the model use it.

What it is

A server can expose three kinds of thing:

• Tools: actions the model can call, each with an input schema.
• Resources: data the application can read.
• Prompts: templates the user can pick.

A local server runs as a subprocess and talks over standard input and output (stdio). A remote server is a web service using the Streamable HTTP transport, usually behind authentication. See the MCP specification.

Why it matters

One server works with every MCP client, so a service publishes a single integration instead of one per AI tool. A server is also an entry point into your systems: what its tools can do, and which keys they act with, needs the same care as any API.

In AI CodeOps

AI CodeOps runs a remote MCP server at https://console.aicodeops.dev/mcp, so Claude Code, Cursor or any client that supports Streamable HTTP can work with your workspace. It has six tools:

list_projects        check_release_gate
start_session        list_findings
get_session          get_evidence_pack

Each call acts as the key's creator, through the same role checks as the console. start_session and get_evidence_pack also need their permission listed on the key's scope. The server is stateless and accepts 120 requests a minute from one address. AI CodeOps is also an MCP client: see MCP servers for connecting your own.