Documentation Lookup

Retrieve current, version-specific documentation and code examples for any library via Context7 MCP tools (primary) or CLI (fallback).

Source: github.com/upstash/context7 · npm: @upstash/context7-mcp Skill updated: 2026-03-22

When to Use

Activate automatically (no explicit "use context7" needed) when:

• Looking up API endpoints, classes, functions, or method parameters
• Checking configuration options or CLI commands
• Generating code that uses a specific library or service
• Debugging issues related to frameworks, SDKs, or APIs
• Retrieving setup instructions, examples, or migration guides
• Verifying version-specific behavior or breaking changes

When NOT to Use

• Pure logic / algorithm questions with no external library
• Questions about project-internal code (use codebase search instead)
• When the user explicitly says not to use external docs

Workflow (MCP — primary)

Two-step process: resolve the library → query the docs.

Step 1: Resolve the Library ID

Call resolve-library-id with:

• libraryName: the library name (e.g., "next.js", "prisma")
• query: the user's full question (improves relevance ranking)

Each result includes:

• Library ID — Context7-compatible identifier (/org/project)
• Name — library or package name
• Description — short summary
• Code Snippets — number of available code examples
• Source Reputation — authority indicator (High, Medium, Low, Unknown)
• Benchmark Score — quality indicator (100 is the highest)
• Versions — list of available versions (/org/project/version)

Select the best match based on:

1. Exact name match (prioritized)
2. Higher benchmark score (max 100)
3. Source reputation: High > Medium > Low
4. More code snippets = better documentation coverage
5. If user mentions a version, use version-specific ID
6. For ambiguous queries, request clarification before proceeding

Skip this step if the user provides an explicit library ID like /org/project.

Step 2: Query Documentation

Call query-docs with:

• libraryId: the selected ID (e.g., /vercel/next.js)
• query: specific, descriptive question

The output contains code snippets (titled, with language-tagged blocks) and info snippets (prose explanations with breadcrumb context).

Step 3: Use in Response

• Answer using the fetched documentation, not training data
• Include relevant code examples from the docs
• Cite the library version when relevant

Workflow (CLI — fallback)

Use CLI only when MCP tools are unavailable or failing.

# Step 1: Resolve library ID
ctx7 library <name> <query>

# Step 2: Query documentation
ctx7 docs <libraryId> <query>

Run directly without installing: npx ctx7@latest <command>

Rules

• Max 3 calls per question — if not found after 3 attempts, use best result
• Library IDs always with / prefix — /facebook/react, not facebook/react
• Descriptive queries — "How to set up JWT auth in Express.js", not "auth"
• Never include secrets — no API keys, passwords, or credentials in queries
• Don't silently fall back — if Context7 fails, tell the user why

Writing good queries

Quality	Example
Good	"How to set up authentication with JWT in Express.js"
Good	"React useEffect cleanup function with async operations"
Bad	"auth"
Bad	"hooks"

Common Library IDs

Domain	Library	Context7 ID
Frontend	Next.js	/vercel/next.js
Frontend	React	/facebook/react
Frontend	Vue.js	/vuejs/vue
Frontend	Svelte	/sveltejs/svelte
Styling	Tailwind CSS	/tailwindlabs/tailwindcss
Backend	Express	/expressjs/express
Backend	Fastify	/fastify/fastify
Database	Prisma	/prisma/prisma
Database	Drizzle	/drizzle-team/drizzle-orm
BaaS	Supabase	/supabase/supabase
Blockchain	Ethers.js v6	/websites/ethers_v6
Blockchain	Viem	/wevm/viem
Blockchain	Wagmi	/wevm/wagmi
Blockchain	OpenZeppelin	/openzeppelin/contracts
Blockchain	Foundry	/foundry-rs/foundry
Blockchain	Hardhat	/nomicfoundation/hardhat
Wallet	Reown AppKit	/websites/reown
Testing	Vitest	/vitest-dev/vitest
Testing	Playwright	/microsoft/playwright

New Project Setup

When starting a new project or repo, proactively offer to set up Context7 documentation lookup. Do this whenever you detect:

• A new AGENTS.md being created or an existing one without a Context7 section
• A new .agents/workflows/ directory being set up
• The user saying "new project", "new repo", "init", or similar

Setup steps

1. Scan package.json (or Cargo.toml, go.mod, requirements.txt, etc.) to identify the project's dependencies
2. Resolve library IDs for each key dependency using resolve-library-id
3. Copy workflow template from workflow-template.md to .agents/workflows/find-docs.md and fill in the library table
4. Add Context7 section to AGENTS.md using the snippet from agents-md-section.md with resolved library IDs
5. Confirm with the user — show the list of detected libraries and their IDs

Templates

Template	Purpose	Copy to
workflow-template.md	Project workflow for /find-docs slash command	.agents/workflows/find-docs.md
agents-md-section.md	Context7 section for AGENTS.md	Append to AGENTS.md

Authentication

Context7 works without authentication. For higher rate limits:

# Option A: environment variable
export CONTEXT7_API_KEY=your_key

# Option B: OAuth login (CLI only)
ctx7 login

Get a free API key at context7.com/dashboard.

Error Handling

Error	Action
Quota exceeded	Inform user, suggest getting API key at context7.com/dashboard
No results	Refine query, try alternative library name
Wrong library match	Be explicit about which library was selected and why
Network error	Fall back to training data, but tell the user

⚠️ Gotchas

• Don't guess library IDs — always resolve first, IDs are not intuitive (e.g., ethers.js is /websites/ethers_v6, not /ethers-io/ethers.js)
• Single-word queries return garbage — always use full questions
• Version-specific IDs have special format — /org/project/version, get available versions from resolve-library-id output
• Rate limits without API key — free tier has low limits, recommend getting a key at context7.com/dashboard
• MCP is preferred over CLI — if MCP tools are available, always use them instead of running ctx7 CLI commands