Context7
Retrieve current documentation for software libraries by querying the Context7 API. This is especially useful when you're unsure about an API's current interface — library docs change frequently and your training data may be outdated.
Requires CONTEXT7_API_KEY environment variable.
When to Use
- Looking up how a library function works (e.g., "how does
useEffectcleanup work?") - Checking if an API has changed in a newer version
- Finding code examples for a specific library feature
- Verifying correct import paths or function signatures
- Installing or configuring a library and needing setup docs
When NOT to Use
- General programming questions (e.g., "how do closures work in JS?")
- Questions about your own project's internal code
- Topics unrelated to a specific open-source library or framework
Workflow
DO NOT read script source code. Run scripts directly and use --help for usage.
Step 1: Search for the Library
python3 scripts/context7.py search <library> <topic>Returns TSV with top 5 matches: id, title, snippets. Use the id from the best-matching row for the fetch step.
Step 2: Fetch Documentation
python3 scripts/context7.py fetch <library_id> <topic> [--max-tokens N]Fetches documentation snippets relevant to the topic, truncated to a token budget (default: 5000).
Choosing --max-tokens:
| Scenario | Tokens | Why |
|---|---|---|
| Quick lookup (one function signature) | 2000 | Keeps output focused, faster response |
| Typical usage (API patterns, examples) | 5000 (default) | Good balance of depth and brevity |
| Broad exploration (migration guide, full API surface) | 8000–10000 | Needed when topic spans multiple sections |
Run python3 scripts/context7.py --help for full usage.
Examples
# Find React library ID, then fetch useState docs
python3 scripts/context7.py search react "useState hook"
python3 scripts/context7.py fetch /websites/react_dev "useState hook with objects"
# Smaller budget for a quick lookup
python3 scripts/context7.py fetch /vercel/next.js "middleware redirect" --max-tokens 2000
# Broader exploration
python3 scripts/context7.py fetch /langchain-ai/langchainjs "retrieval chain setup" --max-tokens 8000Rules
- Write specific queries —
"useState hook with objects"retrieves much better results than"hooks", because the API ranks snippets by relevance to your query. - Always search before fetching — Library IDs aren't guessable (e.g.,
/websites/react_dev), so you need the search step to find the right one. - Match
--max-tokensto the task — Use the table above. Overshooting wastes context window; undershooting may miss the answer.