diegosouzapw Context7 MCP Usage Patterns
v2.88 Key Changes (MODEL-AGNOSTIC)
- Model-agnostic: Uses model configured in
~/.claude/settings.jsonor CLI/env vars - No flags required: Works with the configured default model
- Flexible: Works with GLM-5, Claude, Minimax, or any configured model
- Settings-driven: Model selection via
ANTHROPIC_DEFAULT_*_MODELenv vars
ultrathink - Take a deep breath. We're not here to write code. We're here to make a dent in the universe.
The Vision
Documentation retrieval should be precise, fast, and authoritative.
Your Work, Step by Step
- Identify library: Extract from the user’s request.
- Resolve ID: Use Context7 to find the exact source.
- Query: Ask for targeted, actionable guidance.
- Fallback: Use MiniMax when Context7 lacks coverage.
Ultrathink Principles in Practice
- Think Different: Prefer indexed docs over scraping.
- Obsess Over Details: Ensure the right library ID.
- Plan Like Da Vinci: Define the query before running it.
- Craft, Don't Code: Return only relevant excerpts.
- Iterate Relentlessly: Re-query for clarity.
- Simplify Ruthlessly: Avoid unnecessary searches.
Context7 MCP Usage Patterns
Overview
Context7 MCP provides indexed documentation for popular libraries and frameworks. It's more efficient than web scraping because it uses pre-indexed, structured documentation.
Available Tools
| Tool | Purpose | Parameters |
|---|---|---|
mcp__plugin_context7_context7__resolve-library-id |
Find Context7 library ID | libraryName, query |
mcp__plugin_context7_context7__query-docs |
Query documentation | libraryId, query |
Usage Pattern
# Step 1: Resolve library ID
mcp__plugin_context7_context7__resolve-library-id:
libraryName: "React" # Extract from user query
query: "useTransition hook usage" # Full query for ranking
# Step 2: Query docs with resolved ID
mcp__plugin_context7_context7__query-docs:
libraryId: "/vercel/next.js" # From step 1
query: "How to use useTransition hook"
Decision Tree
Is this about a library/framework?
|
+-- YES --> Is it in Context7?
| |
| +-- YES --> Use Context7 MCP
| | 1. resolve-library-id
| | 2. query-docs
| |
| +-- NO --> Fallback to MiniMax MCP
|
+-- NO --> Use WebSearch (native) or MiniMax MCP
Supported Libraries (Examples)
Frontend
- React (
/facebook/react) - Next.js (
/vercel/next.js) - Vue.js (
/vuejs/vue) - Angular (
/angular/angular) - Svelte (
/sveltejs/svelte)
Languages
- TypeScript (
/microsoft/TypeScript) - JavaScript (MDN)
Backend
- Node.js (
/nodejs/node) - Express (
/expressjs/express) - Fastify (
/fastify/fastify)
CSS/UI
- Tailwind CSS (
/tailwindlabs/tailwindcss) - Chakra UI (
/chakra-ui/chakra-ui)
Databases
- PostgreSQL
- MongoDB (
/mongodb/docs) - Redis
Cost Optimization
| Approach | Token Usage | Quality |
|---|---|---|
| Context7 | ~50% less | High (official docs) |
| Web Search | Baseline | Variable |
| MiniMax | Baseline | High |
Why Context7 saves tokens:
- Pre-indexed documentation
- Structured responses
- No web scraping overhead
- Focused, relevant content
Integration with Ralph Loop
# CLI usage
ralph library "React 19 useTransition"
ralph lib "Next.js 15 app router"
ralph docs "TypeScript generics"
# Slash command
/library-docs React hooks best practices
Fallback Strategy
If Context7 doesn't have the library:
- Log warning: "Library not found in Context7"
- Fallback to
mcp__MiniMax__web_search - Return results from MiniMax
Best Practices
- Extract library name first - Parse user query to identify the library
- Use full query for ranking - Pass complete query to resolve-library-id
- Handle not-found gracefully - Always have MiniMax fallback ready
- Combine with code examples - Request code snippets in your prompt