The agent returns valid-looking code that uses a function missing from my installed package — no error appears. Why?
This is the silent failure of documentation retrieval: wrong library identity or version drift. The resolver found a similarly named package, or query-docs used the latest corpus while your lockfile pins an older release. The returned documentation is internally correct but externally incompatible. The engineering fix: make resolution mandatory, pass the exact version from package.json or the lockfile, and require the agent to echo the resolved library ID and version before writing code. If those identifiers are absent, reject the answer as ungrounded.
Does Context7 replace web search?
No. Context7 is optimized for structured library documentation and code examples; Brave or Exa are better for release announcements, issue discussions, benchmarks, and arbitrary web sources. Use Context7 for what an API is and search for why a production edge case exists.
Why can documentation retrieval still produce a bad implementation?
Documentation describes supported behavior, not your application constraints. The agent can retrieve a correct API example and still violate your auth model, database boundaries, or framework conventions. Pair documentation grounding with repository context and tests.
Can it retrieve private internal libraries?
The public corpus is its primary strength. Private documentation requires a separate internal retrieval layer or a local MCP server; never paste proprietary source into a public documentation query.