Skills Mcp

ALWAYS CALL THIS FIRST — before any other skills_ tool, no exceptions. This is the ONLY valid source of skill_ids. Do NOT use skill_ids from skills_list_all without first running this search to verify relevance. Returns ranked results with similarity scores. Required next actions: score > 0.6 → call skills_get_body with that skill_id score 0.4–0.6 → read the description, then decide whether to load score < 0.4 → no match, proceed without a skill Query tips: be specific about the task. GOOD: 'implement Stripe webhook signature verification in Python' BAD: 'stripe' — too vague, produces poor matches.

BROWSING — See all available skills without semantic search. Use when you want to explore the full registry by category or tag. WARNING: skill_ids returned here have NOT been scored for relevance to your current task. After browsing, you MUST still call skills_find_relevant to verify relevance before loading any skill with skills_get_body. Do NOT pass skill_ids from this listing directly to skills_get_body. Returns lightweight frontmatter (skill_id, name, tags, complexity_level, has_tier3) to keep token usage reasonable. Supports pagination: use offset to skip results, limit to control batch size.

Load full skill instructions. Call ONLY after skills_find_relevant returns this skill_id with score > 0.6. REQUIRED PREREQUISITES — both must be true before calling this tool: 1. You called skills_find_relevant with a specific query 2. This skill_id appeared in those results with score > 0.6 If either prerequisite is missing, STOP and call skills_find_relevant first. Calling this tool with an unverified skill_id (e.g. one from skills_list_all or a guessed ID) will load a skill that may be completely irrelevant to your task, wasting tokens and producing wrong output. Returns: - instructions: step-by-step expert guidance — read and follow entirely - tier3_manifest: available reference files, scripts, assets After loading: follow the instructions precisely. Call Tier-3 tools ONLY when the instructions explicitly name a specific file.

OPTIONAL — Load config variants and constraints for a skill. REQUIRED PREREQUISITES — all must be true before calling this tool: 1. You called skills_find_relevant and got a score > 0.6 2. The skill_id came from those search results 3. The user asked about configuration, or skills_get_body mentioned options Do NOT call by default. Most tasks complete with skills_get_body alone.

Fetch a reference document bundled with a skill. STRICT PREREQUISITES — ALL THREE must be true before calling this tool: 1. You called skills_find_relevant and received a score > 0.6 2. You called skills_get_body and read the full instructions 3. Those instructions explicitly say 'see FILENAME.md' or 'refer to FILENAME' If any prerequisite is missing, do NOT call this tool. Speculative loading wastes tokens and does not improve task quality. Pass filename='list' to see available reference files for a skill.

Execute a helper script bundled with a skill. STRICT PREREQUISITES — ALL THREE must be true before calling this tool: 1. You called skills_find_relevant and received a score > 0.6 2. You called skills_get_body and read the full instructions 3. Those instructions explicitly direct you to run a specific script file If any prerequisite is missing, do NOT call this tool. Script source is NEVER returned — only stdout, stderr, and exit_code. Scripts run sandboxed in an isolated temp directory with a 30-second timeout. Deployment note: script execution requires the local server. The Cloudflare Workers deployment returns the manifest only. Two-phase use: 1. filename='list' → see available scripts and descriptions 2. filename='<script.py>' + optional input_data → execute

Fetch a template or static resource bundled with a skill (markdown templates, config starters, example data files). STRICT PREREQUISITES — ALL THREE must be true before calling this tool: 1. You called skills_find_relevant and received a score > 0.6 2. You called skills_get_body and read the full instructions 3. Those instructions explicitly reference a specific asset file If any prerequisite is missing, do NOT call this tool. Two-phase use: 1. filename='list' → see the full asset manifest 2. filename='<file>' → fetch content (use as a starting template, adapt to task)