# Documentation Lookup > Fetch current library, framework, and API documentation via the Context7 MCP instead of relying on training data. Use this skill whenever the user asks a question that depends on accurate, up-to-date behavior of any library, framework, SDK, or API — including setup/configuration questions, code examples, API references, migration guides, or version-specific behavior. Trigger even if the user doesn't name the skill explicitly but clearly needs current docs to get a correct answer. Use when this capability is needed. - Skill: `tomevault-io/documentation-lookup-4` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/documentation-lookup-4` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/documentation-lookup-4/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/documentation-lookup-4 --- # Documentation Lookup (Context7) Answer library/framework/API questions using live documentation from the Context7 MCP, not training data. Two tools are available: **resolve-library-id** (find the Context7 library ID) and **query-docs** (fetch docs for that ID). ## Workflow ### 1. Resolve the library ID Call **resolve-library-id** with: - **libraryName**: the library or product name from the user's question. - **query**: the user's full question (improves relevance ranking). This returns candidate library IDs. Pick the best match by considering exact name match, benchmark score (higher = better docs quality), source reputation, and version alignment if the user specified one. A valid library ID is required before calling query-docs. If resolution returns no results, tell the user the library wasn't found in Context7 and offer to help via web search or training knowledge instead. ### 2. Fetch the documentation Call **query-docs** with: - **libraryId**: the ID selected in step 1. - **query**: the user's specific question. Be precise — vague queries return vague results. If the returned content doesn't answer the question, try one more query-docs call with a rephrased question before falling back. State uncertainty clearly rather than guessing. ### 3. Answer using the fetched docs Respond using the current documentation. Include relevant code examples from the docs when helpful, and cite the library version when it matters. ## Constraints - **Max 3 MCP calls per question** (combined resolve + query). Each call adds latency and token cost; keep it efficient. If 3 calls aren't enough, use the best information gathered so far and flag gaps honestly. - **No secrets in queries**: strip API keys, passwords, tokens, and other credentials from any text sent to resolve-library-id or query-docs. ## Failure handling - **No resolution results**: the library may not be indexed in Context7. Say so and offer alternative approaches (web search, training knowledge). - **Irrelevant docs returned**: rephrase the query once. If still off-target, answer with what you have and note the limitation. - **User didn't specify a library**: if the question implies a library dependency but doesn't name one, ask which library they're using before calling resolve. ## Example User asks how to configure middleware in a web framework. 1. Call **resolve-library-id** with the framework name and the user's question. 2. From results, pick the official library ID (best name match + highest benchmark score). If the user mentioned a version, prefer a version-specific ID if available. 3. Call **query-docs** with that ID and the question. 4. Answer using the returned docs, including a minimal code example if the docs provide one. --- > Source: [prisma-ddti/skills](https://github.com/prisma-ddti/skills) — distributed by [TomeVault](https://tomevault.io).