MCP Integration

grepai includes a built-in MCP (Model Context Protocol) server that allows AI agents to use semantic code search as a native tool.

What is MCP?

Model Context Protocol (MCP) is an open standard for AI tool integration, supported by:

Native tool access: AI models see grepai as a first-class tool, not a shell command
Subagent inheritance: MCP tools are automatically available to subagents
Structured data: JSON responses by default, no parsing required
Tool discovery: MCP tools are automatically discovered by AI models

Tool	Description	Parameters
`grepai_search`	Semantic code search	`query` (required), `limit` (default: 10)
`grepai_trace_callers`	Find callers of a symbol	`symbol` (required)
`grepai_trace_callees`	Find callees of a symbol	`symbol` (required)
`grepai_trace_graph`	Build complete call graph	`symbol` (required), `depth` (default: 2)
`grepai_index_status`	Check index health	none

Use the claude mcp add command to register grepai as an MCP server:

claude mcp add grepai -- grepai mcp-serve

This automatically configures grepai in your Claude Code settings.

Add to .cursor/mcp.json in your project:

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

Add to your Windsurf MCP configuration:

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

Once configured, AI agents can use grepai tools directly:

Search example:

Tool: grepai_search
Arguments: {"query": "user authentication flow", "limit": 5}

Trace callers example:

Tool: grepai_trace_callers
Arguments: {"symbol": "HandleLogin"}

Index status example:

Tool: grepai_index_status
Arguments: {}

Before using MCP mode, ensure: