SkillMCP
Serves project-specific skills and behavioral rules to AI agents via MCP, enabling automatic injection of behavioral rules and on-demand knowledge for coding assistants like Claude Code and Gemini CLI.
README
SkillMCP
Serves project-specific skills and behavioral rules to AI agents via MCP.
Works with Claude Code, Gemini CLI, Cursor, and Antigravity.
Pairs well with LearnSkill (behavioral auditing) and claude-mem (long-term memory).
Quick Start
cd /path/to/project
skills-mcp init . # scaffold .agents/ skills
Restart your agent host to pick up the new skills.
How it works
Injects knowledge into every agent session automatically via the MCP instruction block.
AGENT.md — behavioral rules
Markdown files injected into the system prompt at session start. All sources are combined; none are dropped.
| Source | Location |
|---|---|
| Bundled (SkillMCP install) | <skillmcp>/.agents/AGENT.md |
| Configured agent folders | AGENT.md in each agent_folders entry |
Skills — on-demand knowledge
Markdown skill files the agent fetches with list_skills / read_skill as needed. Later entries in agent_folders win on name collision.
| Source | Location |
|---|---|
| Bundled (SkillMCP install) | <skillmcp>/.agents/skills/ |
| Configured agent folders | skills/ subdir of each agent_folders entry |
Setup
-
Install
cd SkillMCP && uv sync -
Initialize
skills-mcp init . -
Configure Edit
skillmcp.tomlto add agent folders. Last entry wins on collision.agent_folders = [ "/path/to/shared/agents", ".agents/", ]Each agent folder can contain:
AGENT.md— behavioral rules injected into every session (all folders combined)skills/— skill library scanned forlist_skills/read_skill
Telemetry & Metrics
SkillMCP automatically tracks usage metrics to build a scalable dataset of agent behavior and skill utilization. Telemetry data is persisted to telemetry.json in your project root.
Telemetry Dataset (telemetry.json)
Tracks sessions, tool counts, and skill access leaderboards:
{
"TotalSessions": 100,
"TotalSkillCalls": 5,
"ToolCalls": {
"verify_setup": 10,
"list_skills": 12,
"read_skill": 5,
"skill_health": 1
},
"Skills": [
{ "role-plan": 3 },
{ "role-research": 2 }
]
}
Health Diagnostics (skill_health)
The server exposes a health-check tool skill_health that returns details about the server health and the sequence number of the current tool execution:
{
"status": "healthy",
"call_number": 5,
"total_sessions": 100,
"total_skill_calls": 5,
"checked_at": "2026-05-17T16:00:00Z"
}
CLI Reference
| Command | Description |
|---|---|
init [path] |
Scaffold .agents/, skillmcp.toml, AGENT.md, register MCP |
doctor |
Verify directory layout and MCP registration |
mcp register |
Re-register with all agent hosts (Claude, Gemini, etc.) |
Troubleshooting
- Stale Paths: If you move your project, run
skills-mcp mcp registerfrom the new location to update absolute paths in the MCP configs. - Missing Skills: Run
skills-mcp doctorto see exactly whichskillmcp.tomlis being discovered and how many skills were found.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。