research

Research

Source-grounded docs research for implementation details that may be stale in training data.

Strategy (priority order)

Try these in order and stop at the first source that is authoritative, current enough for the requested version, and contains the API/config detail needed to answer.

1. Context7 MCP — best for library/framework/SDK documentation

mcp__context7__resolve-library-id  →  mcp__context7__query-docs

Prefer this for npm packages, Python libraries, and popular frameworks.

2. GitHub CLI — for repos not indexed by Context7

gh repo view <owner>/<repo> --json description,readme
gh api repos/<owner>/<repo>/contents/docs

3. WebFetch → WebSearch — for everything else

• WebFetch a specific URL if you have it
• WebSearch when you need to find the right page first

If a tool is unavailable, skip to the next available source and say which source could not be checked.

Completion Gate

Stop only when:

• the source is authoritative for the library/framework/API
• the source is current enough for the requested version
• the source contains the API syntax, config option, migration note, or example needed to answer
• unavailable tools or missing version information are disclosed

Output Format

Deliver actionable implementation guidance — not a raw docs dump.

Key patterns — show working code examples for the most common use cases

Common pitfalls — what trips people up, especially on first use

Version notes — call out anything version-specific, especially for fast-moving ecosystems (Next.js App Router vs Pages, React 19 changes, Python 3.12+ features)

Authoritative sources — link to the official docs page you drew from so the user can verify

When to Search vs When to Answer From Memory

Search when:

• The user is actively writing code that uses a library
• The question involves specific API syntax, config options, or method signatures
• The library had major changes in the past 12 months (anything in the JS/Python ecosystem)

Answer from memory when:

• The question is about a fundamental concept (e.g., "what is a closure")
• The user is asking for a high-level explanation, not implementation details
• The library is very stable and the question is about basic usage (e.g., standard library functions)