Showing 28 verified skills. 284 preview entries are hidden until we confirm a real source. Show preview skills · Why?
Create custom tool integrations that extend any AI assistant
claude install anthropics/skills/mcp-builderComplete toolkit for authoring MCP servers with proper schema definitions, error handling, and transport configuration.
This is the actual SKILL.md file that powers this skill. Copy it to install.
---
name: mcp-builder
description: |
Trigger when the user wants to build an MCP server, create MCP tools,
or extend AI capabilities with custom integrations. Phrases: "build MCP",
"create MCP server", "custom tools", "MCP integration".
allowed-tools:
- Bash(command *)
- Read
- Write
- Edit
- Glob
- Grep
---
# MCP Builder
Build Model Context Protocol servers from scratch with tools, resources,
and transport configuration.
## Prerequisites
- Node.js 18+ or Python 3.10+
- `npm install @modelcontextprotocol/sdk` (TypeScript)
- or `pip install mcp` (Python)
## Steps
1. **Decide on tools and resources.** List what your MCP server will expose:
- Tools: actions the AI can take (e.g., run query, send message)
- Resources: data the AI can read (e.g., database schema, config files)
2. **Scaffold the server (TypeScript):**
```typescript
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
const server = new McpServer({
name: "my-server",
version: "1.0.0",
});
server.tool("my_tool", "Description of what it does", {
param: { type: "string", description: "What this param is" },
}, async ({ param }) => {
// Implementation
return { content: [{ type: "text", text: "Result" }] };
});
const transport = new StdioServerTransport();
await server.connect(transport);
```
3. **Add to MCP config:**
```json
{
"mcpServers": {
"my-server": {
"command": "node",
"args": ["path/to/server.js"]
}
}
}
```
4. **Test the server:**
- Run it manually and send test requests
- Verify tool schemas are correct
- Check error handling for invalid inputs
5. **Best practices:**
- Always validate inputs with schemas
- Return structured errors, not raw exceptions
- Keep tools focused - one action per tool
- Use descriptive names and descriptions (the AI reads them)
mkdir -p ~/.claude/skills/mcp-builder~/.claude/skills/mcp-builder/SKILL.mdResulting file structure:
~/.claude/
skills/
mcp-builder/
SKILL.md <-- skill definitionSkills are loaded automatically by Claude Code when you start a new session. The skill name and description in the frontmatter determine when Claude triggers it.
Recommended from shared domain, career, and tool overlap with MCP Builder
Inspect Redis keys, TTLs, and cache hit rates from your AI editor
Both used by Software Engineer, DevOps Engineer
Query Sentry errors and analyze stack traces from your AI editor
Both used by Software Engineer, DevOps Engineer
Give your AI assistant safe, sandboxed access to your filesystem
Both used by Software Engineer, DevOps Engineer
Manage AWS resources - S3, Lambda, and CloudWatch - from your editor
Both used by Software Engineer, DevOps Engineer
Turn one prompt into a parallel pipeline across Claude, Codex, Gemini, and Cursor agents -- then get a unified result back
Both used by Software Engineer, DevOps Engineer
Query databases, inspect schemas, and explore data from your AI editor
Both used by Software Engineer
MCP Builder