We struggled with context overload using LLMs (Claude Code, Cursor etc) on large APIs. Even with doc search tools (Mintlify MCP, etc.), models would grab too much irrelevant info, miss types/params/schemas, or hallucinate context — so LLM outputs were unreliable.

To solve this, we turned our OpenAPI spec into highly modular docs:

- Every schema, endpoint, or parameter lives at a deterministic URL (/schemas/Product, /operations/CreateOrder, etc.)

- Each doc page has the TypeScript definition, relevant OpenAPI fragment, and links to related items

- LLMs (and human devs!) always get exactly the info they need—no more, no less.

- We automated the flow so docs auto update whenever OpenAPI changes

Result: much more accurate, reliable LLM-powered codegen and improved developer productivity.

Would love to know:

- Has anyone else hit this context overload problem?

- Are modular, LLM-optimized API docs for each API entity something others want/need?

Happy to share our approach or hear about others’ solutions!