| name | context7-mcp |
| description | Fetch current library documentation, API references, and code examples for any developer technology (React, tokio, wgpu, Prisma, Supabase, Express, Tailwind, Django, Spring Boot, …) — even well-known ones. Use this skill whenever the user asks setup or configuration questions ("How do I configure Axum middleware?"), requests code involving libraries ("Write a Prisma query for…"), needs API references ("What are the Supabase auth methods?"), or mentions a specific framework or crate by name. Use even when you think you know the answer — training data may be outdated. Source — upstream `upstash/context7` (MIT), `skills/context7-mcp/SKILL.md`, imported 2026-05-19 and rewired onto the native Rust `mcp__aphrody__context7_*` tools. |
| license | MIT |
| upstream | https://github.com/upstash/context7/tree/master/skills/context7-mcp |
Context7 (via aphrody MCP)
Mode /goal permanent : décider seul, ne pas s'arrêter avant réponse complète.
When the user asks about libraries, frameworks, or needs code examples,
use Context7 — wired into this plugin as two native Rust tools on
the unified aphrody-mcp stdio server — to fetch current documentation
instead of relying on training data.
Wired Tools
| Tool | Purpose |
|---|
mcp__aphrody__context7_resolve_library_id | Resolve a package / framework name to a Context7 library ID |
mcp__aphrody__context7_query_docs | Pull current docs + code snippets for a resolved library ID |
The two tools call https://mcp.context7.com/api/v2/{libs/search,context}
in pure Rust (reqwest + rustls/ring). They use the CONTEXT7_API_KEY
env var as a Bearer token if present — the public tier works
unauthenticated.
When to Use This Skill
Activate this skill when the user :
- Asks setup or configuration questions ("How do I configure Axum middleware?")
- Requests code involving libraries ("Write a Prisma query for …")
- Needs API references ("What are the Supabase auth methods ?")
- Mentions specific frameworks (React, Vue, Svelte, Express, Tailwind, …)
- Mentions Rust crates by name (tokio, reqwest, wgpu, serde, …) — Context7
indexes docs.rs in addition to JS / Python / etc.
How to Fetch Documentation
Step 1 — Resolve the Library ID
Call mcp__aphrody__context7_resolve_library_id with :
libraryName : The library name extracted from the user's question
(e.g. "React", "tokio", "Prisma").query : The user's full question (improves relevance ranking).
Step 2 — Select the Best Match
From the resolution results, choose based on :
- Exact or closest name match to what the user asked for.
- Higher benchmark scores indicate better documentation quality.
- If the user mentioned a version (e.g. "React 19"), prefer
version-specific IDs (format
/org/project/version).
Step 3 — Fetch the Documentation
Call mcp__aphrody__context7_query_docs with :
libraryId : The selected Context7 library ID (e.g. /facebook/react).query : The user's specific question.
Step 4 — Use the Documentation
Incorporate the fetched documentation into your response :
- Answer the user's question using current, accurate information.
- Include relevant code examples from the docs.
- Cite the library version when relevant.
Guidelines
- Be specific : pass the user's full question as the query for
better results.
- Version awareness : when users mention versions ("React 19"), use version-specific library IDs if available from the
resolution step.
- Prefer official sources : when multiple matches exist, prefer
official / primary packages over community forks.
- Library IDs require a
/ prefix — /facebook/react, not
facebook/react.
- Always resolve first — calling
context7_query_docs with a
non-resolved name will return a structured error envelope
(CONTEXT7_BAD_REQUEST).
Error Handling
The tools return a structured JSON error envelope on failure :
{
"error": "CONTEXT7_TIMEOUT" | "CONTEXT7_UNAVAILABLE" | "CONTEXT7_BAD_REQUEST",
"reason": "<details>",
"endpoint": "https://mcp.context7.com/api/v2/..."
}
If the envelope reports CONTEXT7_BAD_REQUEST with a Monthly quota reached snippet, tell the user the public quota is exhausted and
recommend setting CONTEXT7_API_KEY (from
context7.com/dashboard) in
~/.bashrc / Windows env vars. Do not silently fall back to training
data — always tell the user why Context7 was not used.
Comparison with other tools in this plugin
| Task | Preferred tool |
|---|
| Library docs / API reference | context7_query_docs (this skill) — versioned, indexed, code snippets |
| Microsoft-specific docs | mcp__aphrody__microsoft_docs_search (Azure / .NET / Windows / M365) |
| Generic web page → Markdown | mcp__aphrody__universal_web_fetch (any URL, Jina reader-proxy via Google Style Guide tool) |
