Your MCP server's real constraint is the context window, not the API

Short summary

When building an MCP server in front of a large document corpus, the real constraint is the context window, not the API. A naive tool returning 40 meeting transcripts would emit 325K tokens — impossible for a 200K window. The solution: the Worker does the reading and returns only matching passages, with a single output budget enforced across the whole run. The response must distinguish cuts for size from cuts for relevance.

• •Context window arithmetic, not API limits, is the primary design constraint for MCP tools over large corpora
• •Server-side scanning returns excerpts (~19K tokens) instead of full transcripts (~325K tokens)
• •A single output budget across the whole run prevents independent per-limit caps from overrunning the context window

Generated with AI, which can make mistakes.