grepai-mcp-claude（grepai MCP Claude 搜索）

GrepAI MCP Integration with Claude Code

This skill covers integrating GrepAI with Claude Code using the Model Context Protocol (MCP).

When to Use This Skill

• Setting up GrepAI in Claude Code
• Enabling semantic search for AI coding assistant
• Configuring MCP server for Claude
• Troubleshooting Claude Code integration

What is MCP?

Model Context Protocol (MCP) allows AI assistants to use external tools. GrepAI provides an MCP server that gives Claude Code:

• Semantic code search
• Call graph analysis
• Index status monitoring

Prerequisites

1. GrepAI installed
2. Ollama running (or other embedding provider)
3. Project indexed (grepai watch)
4. Claude Code installed

Quick Setup

One command to add GrepAI to Claude Code:

claude mcp add grepai -- grepai mcp-serve

That's it! Claude Code can now use GrepAI tools.

Manual Configuration

If you prefer manual setup, add to Claude Code's MCP config:

Location

• macOS/Linux: ~/.claude/mcp.json
• Windows: %APPDATA%\Claude\mcp.json

Configuration

{
  "mcpServers": {
    "grepai": {
      "command": "grepai",
      "args": ["mcp-serve"]
    }
  }
}

With Working Directory

If you want GrepAI to always use a specific project:

{
  "mcpServers": {
    "grepai": {
      "command": "grepai",
      "args": ["mcp-serve"],
      "cwd": "/path/to/your/project"
    }
  }
}

Verifying Installation

Check MCP Server

# Start MCP server manually to test
grepai mcp-serve

You should see:

GrepAI MCP Server started
Listening for requests...

In Claude Code

Ask Claude:

"Search the codebase for authentication code"

Claude should use the grepai_search tool.

Available Tools

Once connected, Claude Code has access to these tools:

Tool Usage Examples

Semantic Search

Claude request:

"Find code related to user authentication"

Claude uses:

{
  "tool": "grepai_search",
  "parameters": {
    "query": "user authentication",
    "limit": 5,
    "compact": true
  }
}

Trace Analysis

Claude request:

"What functions call the Login function?"

Claude uses:

{
  "tool": "grepai_trace_callers",
  "parameters": {
    "symbol": "Login",
    "compact": true
  }
}

Index Status

Claude request:

"Is the code index up to date?"

Claude uses:

{
  "tool": "grepai_index_status",
  "parameters": {
    "verbose": true
  }
}

Compact Mode

By default, MCP tools return compact JSON to minimize tokens:

{
  "q": "authentication",
  "r": [
    {"s": 0.92, "f": "src/auth/middleware.go", "l": "15-45"},
    {"s": 0.85, "f": "src/auth/jwt.go", "l": "23-55"}
  ],
  "t": 2
}

This reduces token usage by ~80% compared to full content.

Working Directory

The MCP server uses the current working directory. Ensure:

1. GrepAI is initialized in your project
2. Index exists (run grepai watch first)
3. Start Claude Code from your project directory

Option 1: Start Claude from Project Directory

cd /path/to/your/project
claude  # Claude Code now uses this directory

Option 2: Configure CWD in MCP Config

{
  "mcpServers": {
    "grepai": {
      "command": "grepai",
      "args": ["mcp-serve"],
      "cwd": "/path/to/your/project"
    }
  }
}

Multiple Projects

For multiple projects, you can:

Option 1: Multiple MCP Servers

{
  "mcpServers": {
    "grepai-frontend": {
      "command": "grepai",
      "args": ["mcp-serve"],
      "cwd": "/path/to/frontend"
    },
    "grepai-backend": {
      "command": "grepai",
      "args": ["mcp-serve"],
      "cwd": "/path/to/backend"
    }
  }
}

Option 2: Use Workspaces

grepai workspace create my-workspace
grepai workspace add my-workspace /path/to/frontend
grepai workspace add my-workspace /path/to/backend

{
  "mcpServers": {
    "grepai": {
      "command": "grepai",
      "args": ["mcp-serve", "--workspace", "my-workspace"]
    }
  }
}

Troubleshooting

Tool Not Available

❌ Problem: Claude doesn't see GrepAI tools

✅ Solutions:

1. Restart Claude Code after config changes
2. Check MCP config syntax (valid JSON)
3. Verify grepai is in PATH
4. Test: grepai mcp-serve manually

Search Returns No Results

❌ Problem: Searches return empty

✅ Solutions:

1. Ensure grepai watch has run
2. Check working directory has .grepai/
3. Verify index exists: grepai status

Connection Refused

❌ Problem: MCP server won't start

✅ Solutions:

1. Check Ollama is running: curl http://localhost:11434/api/tags
2. Verify config: cat.grepai/config.yaml
3. Run grepai mcp-serve manually to see errors

Wrong Project Indexed

❌ Problem: Results from wrong codebase

✅ Solutions:

1. Check cwd in MCP config
2. Start Claude from correct directory
3. Verify with grepai_index_status tool

Best Practices

1. Keep index updated: Run grepai watch --background
2. Use compact mode: Reduces token usage
3. Set working directory: Explicit cwd in config
4. Check status first: Use grepai_index_status
5. Restart after config: Claude needs restart for MCP changes

Removing Integration

To remove GrepAI from Claude Code:

claude mcp remove grepai

Or manually edit ~/.claude/mcp.json and remove the grepai entry.

Output Format

Successful MCP setup:

✅ GrepAI MCP Integration Configured

   Claude Code: ~/.claude/mcp.json
   Server: grepai mcp-serve
   Status: Connected

   Available tools:
   - grepai_search (semantic code search)
   - grepai_trace_callers (find callers)
   - grepai_trace_callees (find callees)
   - grepai_trace_graph (call graphs)
   - grepai_index_status (index health)

   Claude can now search your code semantically!