How MCP servers add to your agent's context
Every MCP server you enable gives your coding agent tools — and every tool has a description and schema the model has to read. Here's how that adds up and how to keep it down.
Checked against sources on 2026-09-19
What an MCP server is
The Model Context Protocol is an open standard for connecting AI applications to outside systems — files, databases, issue trackers, browsers. An MCP server exposes a set of tools; a client such as Claude Code or Cursor connects to it and lets the model call those tools.
Why tools cost context
For the model to use a tool, it has to know the tool exists. Clients ask each server for its tool list, and every tool comes back with a name, a description and a JSON schema of its inputs. Those definitions are given to the model — so a server with dozens of tools adds dozens of descriptions to what the model reads.
Tool results cost context too. Claude Code, for example, warns when a single MCP tool result passes 10,000 tokens and caps it at 25,000 by default.
How clients are reducing it
Claude Code now defers MCP tools by default: at the start of a session it loads only tool names and server instructions, and fetches a tool's full definition when it needs it. Deferral is switched off in some setups — for example with a non-Anthropic API endpoint or older models — in which case definitions load up front.
Other clients handle this differently, so check your client's docs before assuming tools are free to leave enabled.
Keeping it down
Turn off servers you are not using (in Claude Code, with /mcp) and check what is taking up space (with /context). Anthropic's cost guide also suggests preferring command-line tools like gh or aws where they do the job, since they are more context-efficient than an MCP server.