Agents and protocols

Context window

The total amount of text a model can consider at once, including everything the tools put there.

Every connected server contributes its tool descriptions to that budget before the work starts. A setup with many servers can spend a meaningful share of the window describing tools it will never call.

It also has a security consequence that is easy to miss: everything in the window has the model's attention, including a description from a server you connected months ago and have not read since.

Why it matters

It is the reason connecting more servers can make an agent worse rather than better, at no point producing an error message.

The mistake everyone makes

Treating the window as free space. Cost, latency and the odds of the model picking the wrong tool all rise with what is in it.

Related terms

Tool shadowing

A malicious server describing its tool so persuasively that the model prefers it to the legitimate one.

MCP server

A program that exposes tools and data to a model through the Model Context Protocol.

Metered pricing

Paying per run rather than per month, usually because the maker's own cost scales with use.

Success rate

The share of tasks in our fixed set that a listing completed correctly, measured in the sandbox.

See Context window on a real listing

Every term here shows up in the catalogue next to a real result, with the findings published and the limits stated. Free to browse, no account needed.

Open the catalogue