Skip to content

LearnGenerate & answer

Model Context Protocol

MCP

An open protocol that lets AI clients discover and call tools, read resources and use prompts exposed by servers, over JSON-RPC.

In DocSlurpGlossary entry

On this page

The Model Context Protocol (MCP) is an open standard for connecting AI applications to outside capabilities. A server exposes tools (functions the model may call), resources (data the client can read) and prompts (reusable templates). A client, such as a chat app, an IDE or an agent, connects, lists what the server offers, and lets its model call those tools. Messages are JSON-RPC 2.0, carried over stdio for local servers or Streamable HTTP for remote ones.

Before MCP, connecting three assistants to a document search service meant three integrations, each with its own tool format. With MCP the service writes one server. Any compliant client can then ask “what tools do you have?” (tools/list), get back names, descriptions and JSON Schema arguments, and call search with {"q": "How fast can my assistant get the Learjet to London?"} (tools/call).

MCP standardizes the plumbing around tool calling. It doesn’t change how the model decides to call a tool; it changes how the tool is found, described, authorized and invoked.

The gotcha

An MCP server’s tool descriptions and results go straight into the model’s context, so they are an input channel like any other. A malicious or compromised server can plant instructions in a description or a result. Connect only servers you trust, give each the narrowest credentials it needs, and treat tool output as data, not commands. See prompt injection.

In DocSlurp

In DocSlurp

DocSlurp serves MCP at /v1/mcp over Streamable HTTP, with OAuth or a server API key, exposing listProjects, upload, search, chat, getUploadStatus and estimate tools (core/server/src/routes/mcp.ts, core/server/src/services/mcp-tools.ts). Stdio clients use docslurp mcp-server. See MCP.

Sources

Jump to