MCP Grep

MCP Grep - GitHub Code Search

Search over a million public GitHub repositories for real-world code examples and usage patterns.

Version: 0.1.0 (locked 2026-01-20)

Overview

This skill provides access to grep.app, a GitHub code search service that helps you find actual code implementations across a massive collection of repositories. Unlike keyword search, this tool searches for literal code patterns - the exact text that appears in source files.

Key Features:

• Search literal code patterns (like grep)
• Support for regular expressions
• Filter by language, repository, or file path
• Case-sensitive and whole-word matching options
• Multi-line pattern matching with regex

Tools

⚠️ Always check tools/ for exact parameter names before calling.

Use Cases

• API Usage Examples: Find how developers use unfamiliar APIs or libraries
• Syntax Reference: Check correct syntax, parameters, or configuration
• Best Practices: Discover production-ready implementation patterns
• Integration Examples: See how different libraries work together

Usage

Quick Start

⚠️ Before calling any tool, read its documentation in tools/ to get the exact parameter names.

# List available tools and their schemas
node ./scripts/mcp-caller.mjs list

# Then check tools/<tool-name>.md for parameter details

CLI (Single Call)

Run from this skill directory:

node ./scripts/mcp-caller.mjs call <tool> '<json-args>'
node ./scripts/mcp-caller.mjs resource <uri>
node ./scripts/mcp-caller.mjs prompt <name> '<json-args>'
node ./scripts/mcp-caller.mjs list

Programmatic (Batch/Parallel)

For batch operations or complex logic, use the API module:

// example.mjs - Run from this skill directory
import { callTool, listTools } from './scripts/api.mjs';

// First, check available tools and their schemas
const tools = await listTools();
console.log(tools.map(t => ({ name: t.name, params: Object.keys(t.inputSchema?.properties || {}) })));

// Then call with correct parameters (see tools/*.md for details)
const result = await callTool('toolName', { /* check tools/toolName.md for params */ });

// Parallel calls
const results = await Promise.all([
  callTool('tool1', { /* params from tools/tool1.md */ }),
  callTool('tool2', { /* params from tools/tool2.md */ }),
]);

console.log(JSON.stringify(results, null, 2));

Available API functions:

• callTool(name, args) - Call a tool
• listTools() - List available tools (includes inputSchema with parameter info)
• readResource(uri) - Read a resource
• listResources() - List resources
• getPrompt(name, args) - Get a prompt
• listPrompts() - List prompts
• close() - Close connection (optional cleanup)

Search Tips

Good Search Patterns (Literal Code)

• ✅ useState( - Find React useState calls
• ✅ import React from - Find React imports
• ✅ async function - Find async function declarations
• ✅ (?s)try {.*await - Find try-catch with await (regex)

Bad Search Patterns (Keywords)

• ❌ react tutorial - Not literal code
• ❌ best practices - Too generic
• ❌ how to use - Natural language query

Using Regex

Enable regex mode with useRegexp: true for advanced patterns:

# Multi-line pattern (prefix with (?s))
node ./scripts/mcp-caller.mjs call searchGitHub '{"query":"(?s)useState\\(.*loading","useRegexp":true}'

Configuration

Connection settings are in config.toml:

transport = "http"
url = "https://mcp.grep.app"

No authentication required for basic usage. If API key support is added in the future, use the MCP_GREP_API_KEY environment variable.

Error Handling

If a tool call fails (e.g., "method not found", parameter changes):

1. Try to complete the task using other available tools
2. Inform user: "This skill may be outdated - MCP server API has changed"
3. User needs to run mcp-skill-generator to update this skill

Do not retry failed calls repeatedly - prioritize finding alternatives.

Updating

To update this skill with newer MCP server version:

1. Run mcp-skill-generator
2. Provide this skill's path for update

References

• grep.app - Official web interface
• Vercel Blog: Grep a Million GitHub Repositories via MCP
• MCP Documentation

Limitations

• Searches literal code patterns, not semantic understanding
• Results depend on indexed repositories (may not include very recent commits)
• Regular expressions must be supported by the service's regex engine