Skip to content

mcp-compatibility.md

Latest commit

 

History

History
147 lines (100 loc) · 4.11 KB

mcp-compatibility.md

File metadata and controls

147 lines (100 loc) · 4.11 KB
title MCP Compatibility
layout default
parent Architecture
nav_order 3

MCP Compatibility & Multi-Tool Support

This document outlines codetect's compatibility with MCP (Model Context Protocol) clients and plans for supporting other CLI-based LLM tools.

What is MCP?

Model Context Protocol (MCP) is an open standard developed by Anthropic for connecting AI assistants to external tools and data sources. It provides a standardized way for LLM applications to:

  • Discover available tools
  • Call tools with structured inputs
  • Receive structured outputs

codetect implements an MCP server that exposes code search capabilities via the stdio transport.

Current MCP Client Support

Tool Support Status Notes
Claude Code Fully supported Primary target, tested extensively
Cursor Should work MCP support added late 2024
Cline (VS Code) Should work Full MCP support
Continue Should work Open source IDE extension with MCP
Zed Should work Built-in MCP support

Tested Configurations

  • Claude Code on macOS with stdio transport

Community-Reported Configurations

We welcome reports of codetect working with other MCP clients. Please open an issue or PR to add your configuration.

Configuration for MCP Clients

codetect uses the standard .mcp.json configuration format:

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

Most MCP clients will automatically discover this configuration when placed in the project root.

Client-Specific Setup

Claude Code

cd /path/to/project
codetect init    # Creates .mcp.json
codetect index   # Index symbols
claude              # Start Claude Code

Cursor

Cursor reads .mcp.json from the workspace root. After running codetect init, restart Cursor to pick up the configuration.

Cline / Continue

These VS Code extensions typically read MCP configuration from:

  • .mcp.json in workspace root
  • VS Code settings

Non-MCP Tool Support

Current Status

codetect is currently MCP-only. Tools that don't support MCP cannot directly use it.

Tools Without MCP Support

Tool MCP Support Alternative Approach
OpenAI Codex CLI No Planned: HTTP API mode
Google Gemini CLI No Planned: HTTP API mode
Aider No Planned: CLI mode
GPT-Engineer No Planned: CLI mode

Roadmap: Multi-Tool Support

We plan to expand codetect to support non-MCP tools through additional interfaces:

Phase 1: HTTP API Mode (Planned)

Add an HTTP server mode that exposes the same tools via REST API:

# Start HTTP server
codetect serve --port 8080

# Query via curl
curl -X POST http://localhost:8080/search_keyword \
  -H "Content-Type: application/json" \
  -d '{"query": "func main", "top_k": 5}'

This would enable integration with any tool that can make HTTP requests.

Phase 2: CLI Query Mode (Planned)

Add direct CLI commands for querying:

# Direct CLI usage
codetect search "func main" --limit 5
codetect symbol Server --kind struct
codetect semantic "error handling logic"

This would allow tools to shell out to codetect directly.

Phase 3: Language Server Protocol (Considered)

LSP integration could provide IDE-native symbol search and navigation, complementing the MCP interface.

Contributing

We welcome contributions to expand tool support:

  1. Testing with MCP clients: Report which clients work with codetect
  2. HTTP API implementation: Help build the REST interface
  3. CLI query mode: Help build direct CLI commands
  4. Documentation: Add setup guides for specific tools

See CONTRIBUTING.md for guidelines.

Resources