MCP-skill-library-dynamic
This server is an MCP-based skill library that lets you discover, search, organize, fetch, and manage reusable AI skill guidelines and engineering best practices.
Omni-search skills using semantic expansion, fuzzy matching, weighted scoring, multi-dimensional filters, and pagination (find_skills).
Browse all skills and explore the taxonomy of domains, occupations, categories, and uninitialized skills.
Fetch full skill markdown guidelines and detailed metadata/file breakdowns.
Rescan local skill directories and sync the taxonomy cache without restarting.
Update skill metadata (domain, occupation, category, tags) in SKILL.md frontmatter and the master registry.
Install skills from repositories, URLs, or custom markdown content, and uninstall existing ones.
Manage persistent AI session state with TTL expiration (admin only).
Issue, regenerate, delete, validate, and list OAuth API keys with roles and audit logs (admin only).
View system status, network IPs, recent activity logs, and background PM2 services.
Base tools include list_skills and fetch_skill_rule for lightweight clients.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@MCP-skill-library-dynamic List the available skills and fetch the code-review instructions."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Skill Library MCP Server
An enterprise-grade Model Context Protocol (MCP) server for automatically discovering, indexing, and serving 580+ AI Skill Guidelines and Engineering Best Practices. Features an Omni-Search Multi-Dimensional Taxonomy Engine, Semantic Query Expansion, Fuzzy Matching, Weighted Relevance Scoring, PM2 Daemon Management, an interactive TUI Control Dashboard, OAuth Token Lifecycle Management, and a Persistent AI Session Memory Store.
Quick Start
1. Install and Build
npm install
npm run build2. Launch Interactive Control Dashboard (TUI)
npm run menuNote: Navigate using Arrow Keys (Up / Down) + Enter or press numeric shortcut keys [0-6] directly. Mouse tracking is disabled to guarantee native terminal copy/paste.
Related MCP server: docubridge
IMPORTANT: First-Time Setup and Taxonomy Sync
When setting up the server for the first time, or after adding/cloning new skills, you must instruct the AI to synchronize and organize the taxonomy metadata to enable full Omni-Search capabilities.
Example AI Prompt:
"Please call sync_skills and organize all uninitialized skills using update_skill_metadata."The 7-Step AI Librarian Workflow:
sync_skills: Rescan local skill directories to discover newly added skills immediately without restarting the server.explore_taxonomy: Inspect the current breakdown of Domains, Occupations, Categories, and the uninitialized skills count (noninit_count).find_skills({ category: "noninit", limit: 100 }): Retrieve unclassified skills page by page.Analyze Skill: Identify target domains and occupations based on the skill purpose.
Create or Reuse Taxonomy: Reuse existing domain/occupation names if applicable, or create broad new ones (e.g.
gaming,finance,healthcare).Define Specific Tags: Assign specific technology and problem keywords (e.g.
["ecs", "dots", "burst", "optimization"]).update_skill_metadata: Write updates directly to the localSKILL.mdfrontmatter and the master registry simultaneously.
Omni-Search and Taxonomy Discovery Architecture
Skill Library MCP utilizes a 4-dimensional taxonomy model paired with an intelligent relevance engine:
Dimension | Description | Example Values |
| Target industry or broad domain |
|
| Target role or persona |
|
| Technical classification |
|
| Specific technologies, libraries, or problem areas |
|
3-Layer Search Intelligence:
Semantic Query Expansion: Automatically expands search terms to include domain synonyms and related keywords (e.g. searching
gameexpands to[game, gaming, gamedev, unity, unreal, godot]).Fuzzy String Matching: Tolerates typographical errors and partial spellings using string similarity metrics (e.g.
gamming,postgre,optimizaton).Weighted Scoring and 40% Relative Cutoff:
Dynamically scores candidates (Exact Name: +15, Tag match: +10-12, Domain match: +8, Category match: +4).
Automatically filters out candidates scoring below 40% of the top match score to eliminate noise and save token usage.
MCP Client Configuration
1. Remote HTTP / SSE Mode (Recommended for Antigravity, Gemini IDE, Claude Desktop, Cursor)
Add to your IDE configuration (mcp_config.json):
{
"mcpServers": {
"skill-library": {
"url": "http://localhost:8787/sse",
"headers": {
"Authorization": "Bearer <YOUR_API_TOKEN>"
}
}
}
}(Generate Bearer Tokens from option [3] API Keys and Auth in npm run menu)
2. Local Stdio Mode (Direct Process Spawn)
{
"mcpServers": {
"skill-library": {
"command": "node",
"args": ["<ROOT_DIR>/dist/index.js"],
"env": {
"SKILLS_DIR": "<ROOT_DIR>/.agents/skills",
"SKILL_LIBRARY_DATA_DIR": "<ROOT_DIR>/.data"
}
}
}
}(Replace <ROOT_DIR> with the absolute path to the project directory, e.g. C:/Users/admin/Desktop/skill-library-mcp)
Available MCP Tools and Role Permissions (RBAC)
The server enforces Role-Based Access Control (RBAC):
Category | Tool Name | Permission Role | Description |
Discovery and Search |
| Public AI | Omni-Search with weighted scoring, fuzzy matching, multi-dimensional filters, and pagination. |
| Public AI | Browse skills with pagination ( | |
| Public AI | View taxonomy summary (Domains, Occupations, Categories) and unclassified ( | |
| Public AI | Fetch full markdown guidelines and best practices for a skill. | |
| Public AI | Get detailed file breakdown and metadata for a skill. | |
Management and Sync |
| Public AI | Rescan local folders and rebuild taxonomy cache without server restart. |
| Public AI | Assign Domain, Occupation, Category, and Tags with auto-canonicalization. | |
| Public AI | Install skills from repository, URL, or custom markdown content. | |
| Public AI | Remove a skill from the library. | |
AI Sessions |
| Admin Only | Persist arbitrary JSON session state with TTL expiration. |
| Admin Only | Retrieve stored session data by name. | |
| Admin Only | List all active sessions (auto-purges expired records). | |
| Admin Only | Delete session record. | |
Auth and Keys |
| Admin Only | Issue new API tokens with custom roles and TTL expiration. |
| Admin Only | Re-issue new token for an existing key ID. | |
| Admin Only | Delete an API key permanently. | |
| Admin Only | Verify token validity, expiration, and associated role. | |
| Admin Only | List stored API key metadata and usage audit logs. | |
System Status |
| Admin Only | View network IPs (Tailscale/LAN/Localhost) and recent activity logs. |
| Admin Only | List all registered and active background services (read-only). |
PM2 Background Daemon Management
The project includes built-in PM2 orchestration for running as a background service:
# Start server daemon in background
npm run pm2:start
# View process table, memory usage, and uptime
npm run pm2:status
# Stream realtime logs
npm run pm2:logs
# Restart or stop the daemon
npm run pm2:restart
npm run pm2:stopMandatory AI Agent System Prompt
To ensure AI agents utilize this system effectively, include this protocol in your system prompt or custom rules:
CRITICAL DIRECTIVE: YOUR VERY FIRST ACTION BEFORE DOING ANYTHING ELSE MUST BE TO CALL THE find_skills TOOL.
You are equipped with the Skill Library MCP (Dynamic Skills and Best Practice Registry).
Before analyzing the request, reading files, or planning, you MUST strictly follow this 3-step protocol:
1. Discover (find_skills({ query: "<task_or_framework>", domain: "<domain>", occupation: "<role>" }))
2. Ensure Available and Load Rules (fetch_skill_rule({ skill_name: "<skill_name>" }) or install_skill)
3. Comply and ImplementLicense
MIT
Available Tools
2 toolsfetch_skill_ruleARead-onlyIdempotent
Fetch the full markdown content and nested rule guidelines of a specific skill.
Inputs: skill_name (required). Returns: files, content, skill_name
list_skillsARead-onlyIdempotent
List all available skills/rules in the knowledge base. Supports optional search query for token efficiency.
Inputs: query. Returns: count, skills
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 2 tools
The two tools have completely distinct purposes: one lists available skills, the other fetches a specific skill's content. There is no overlap or ambiguity between them.
Both tools follow the same verb_noun pattern (list_skills, fetch_skill_rule), making the naming predictable and consistent.
With only 2 tools, the set is on the thin side, but it fits the narrow scope of a read-only skill library. It feels borderline rather than clearly inadequate.
For a read-only knowledge base, list and fetch cover the essential operations. The search option on list_skills adds flexibility. Minor gaps like versioning or metadata retrieval are not critical.
Related MCP Connectors
Read-only catalog of prompt packs and rule files: Cursor, CLAUDE.md, AGENTS.md, SKILL.md.
Serve a folder of Markdown notes as an MCP server: hybrid search, reading, and sourced answers.
Agent-native MCP server over the public saagarpatel.dev corpus. Read-only, stateless.
Search, install and submit Markdown skills for AI agents; private Brain files when signed in.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceServes markdown instructions as tools to provide LLMs with team playbooks, coding standards, domain knowledge, and personal workflows through version-controlled files.5 npmMIT
- AlicenseNot gradedqualityDmaintenanceProvides AI assistants with direct access to local Markdown documentation libraries, enabling them to list, read, and search through docs on demand.MIT
- AlicenseBqualityBmaintenanceEnables AI agents to discover, read, search, and install Markdown-based knowledge (rules, skills, workflows) from a local directory via MCP tools.14102 npm1MIT
- AlicenseNot gradedqualityCmaintenanceTurns any folder of Markdown agent skills into a local MCP server, exposing them as tools, prompts, and resources for MCP-compatible clients.MIT