One line away from Claude Code having reliable docs
Your dependencies → specialized doc agents in seconds
npx claude-docsrequires claude code cli
How It Works
- 1.
npx claude-docs— Configures Context7 MCP, installs slash commands, creates base agents - 2.
claude /sync-docs— Scans your dependencies, validates with Context7, generates library agents - 3.
@research-react— Call agents directly
Or use /research to auto-route:
/research "how do I validate webhooks?"
↓ detects stripe → routes to @research-stripe
↓ queries Context7 → docs-backed answer
Base Agents for Efficient Research
| Agent | Purpose | Source |
|---|---|---|
@codebase-locator | Find WHERE code lives | View |
@codebase-analyzer | Understand HOW code works | View |
@codebase-pattern-finder | Find EXAMPLES in the codebase | View |
@web-search-researcher | Research topics not in Context7 | View |
Plus dynamically generated @research-{library} agents based on your dependencies.
Why Pre-Generated Agents
/sync-docs scans your package.json or requirements.txt, validates each library against Context7, and generates dedicated @research-{library} agents with the correct Context7 library IDs baked in. It also creates a personalized /research command that knows all your available agents and routes queries automatically.
Why not query docs on-the-fly?
- Save context — Agent prompts are loaded only when invoked, not on every request
- Fewer MCP calls — Library IDs are resolved once at sync time, not every query
- Better prompts — Each agent has library-specific instructions and topic hints
- Predictable — You can inspect and edit the generated agents in
.claude/agents/
Why Context7
Context7 indexes up-to-date dev-documentation for thousands of libraries.
We use it because it works. Instead of building and maintaining our own documentation scraper, we chose a reliable provider. Less code to maintain, more time to improve the actual agents.
TODO:
PRs welcome → CONTRIBUTING.md