skills-over-mcp
Enables multi-step, opinionated workflows (skills) as MCP resources with lazy loading, encoding team knowledge like branch naming and test procedures alongside tool execution.
README
skills-over-mcp
Skills over MCP: agent workflows as standard MCP resources with a skill:// URI scheme.
npm install
Quick start
# Start the MCP server (port 3001) and web UI (port 3000)
npm run server &
npm run web
Open http://localhost:3000, type what you need ("deploy to production", "review this PR"), and the LLM routes you to the right skill. Steps walk you through the workflow with live tool execution.
How it works

skill://index.json returns the discovery index. skill://git-workflow/SKILL.md returns the full skill. Skills reference tools by name; the UI renders inline tool panels.
Why skills, not just tools?
Discovery model. tools/list returns full schemas for every tool, all loaded into the LLM context at once. There is no tools/read. Resources have a two-step fetch: resources/list returns lightweight metadata, then resources/read loads one resource on demand. Skills use this to get lazy loading without new protocol primitives.
tools/list → [full schema A, full schema B, ...] // all in context
resources/list → [name A, name B, ...] // just the index
resources/read → full content of A // one at a time
But wouldn't tool search or progressive tool loading solve this? It would fix the discovery problem (finding the right tool from hundreds). But skills solve a different problem: process knowledge.
A tool says create_branch(branch_name: string). A skill says:
- Name it
<type>/<ticket>-<slug>, not whatever you want - Always branch from
mainunless the ticket says otherwise - Run
run_testsbefore pushing - If integration tests fail but your change is frontend-only, that's acceptable
- Open a PR using this specific template
That's team knowledge that doesn't fit in a tool schema. Even with perfect tool discovery, the agent still needs to know how your team uses those tools together. Skills encode that multi-step, conditional, opinionated workflow. Tools are the verbs; skills are the playbook.
Skills included
skills/
├── git-workflow/SKILL.md # Branching, commits, PRs
├── code-review/SKILL.md # Structured review + security checklist
│ └── references/security-checklist.md
└── deploy-service/SKILL.md # Pre-flight → staging → canary → monitoring
Deploy
npm run deploy
Deploys to Cloudflare Workers as a self-contained worker with embedded skill content. Live at https://skills-over-mcp.h3manth.com.
Architecture
- MCP Server (
src/server.ts): Streamable HTTP transport, servesskill://resources and tools - Web Bridge (
src/web.ts): HTTP bridge connecting the browser to the MCP server - Worker (
src/worker.ts): Cloudflare Worker bundling everything for edge deployment - UI (
public/index.html): Intent-based runner, AI-powered skill routing via Cloudflare Workers AI
Timeline
- 2025-12-20 Skills and MCP, Better Together blog post
- 2025-12-22 RFC: Remote Agent Skills, URL-based Skill Import
- 2026-02-01 MCP Interest Group formed; experimental repo created
- 2026-04-14 Initial charter formalized by MCP (via SEP-2149)
Related
- Skills Over MCP Spec, SEP-2640
- experimental-ext-skills, the Interest Group repo
- MCP SDK, protocol implementation
License
MIT © Hemanth.HM
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.