A lazy session’s context on the 200K window as the reader loads domains through load_modules · fig. 2.18.3

fig. 2.18.3 · a lazy session’s context on the 200K window as the reader loads domains through load_modules.

Each segment is the measured tokens of that module’s tool definitions (gpt-tokenizer o200k_base over name, description and input schema), as load_modules registers them; a module in two domains counts once, as the server resolves it. Loading everything holds 85,888 tokens: load_modules itself stays in the list, which an eager start at this commit leaves out.

source: fedpipe 413372f, tools/list over stdio in --eager mode mapped to modules by dist/apis/*/index.js; tokens by gpt-tokenizer o200k_base per module · loadModules.data.ts

View in context: Lazy tool loading over MCP → · All figures

Andrew Fribush, software and computer engineer in Austin. I build consequential systems where software meets institutions: finance, healthcare, commerce, government data, and the mechanisms that govern AI agents. Taking new engagements: hello@andrewfribush.com.

fig. 2.18.3: a lazy session’s context on the 200K window as the reader loads domains through load_modules