How to Configure Platform-Specific MCP Installations for code-review-graph

Use the code-review-graph install command with the --platform flag to generate Cursor or Claude Code-specific MCP configuration files automatically.

The code-review-graph CLI provides a dedicated installation workflow that detects your development environment and writes the correct Model Context Protocol (MCP) settings for each supported AI coding tool. When you need to configure only one platform—such as Cursor or Claude Code—you can target that specific editor without affecting other tools.

Using the Install Command with Platform Flags

The install subcommand accepts a --platform argument that restricts configuration generation to a single target. This prevents unnecessary file creation and ensures paths are optimized for the specific tool chain.

Run one of the following commands from your project root:

# Configure only Cursor

code-review-graph install --platform cursor

# Configure only Claude Code

code-review-graph install --platform claude-code

The command probes for the presence of the target tool by checking for standard configuration directories (~/.cursor or ~/.claude) before writing files. If the platform is detected, the tool generates the appropriate JSON descriptors in your project tree according to the implementation in code_review_graph/skills.py::install_platform_configs (lines 505-550).

Generated Configuration Files

Each platform receives a tailored set of configuration files that tell the AI client how to launch and communicate with the code-review-graph MCP server.

Cursor MCP Configuration

For Cursor, the installer creates a single file at .cursor/mcp.json. This JSON object contains the launch descriptor that Cursor uses to start the MCP server in stdio mode:

• cmd: The executable command (code-review-graph)
• args: The arguments array (["serve"])
• cwd: The working directory pointing to your repository root

You can inspect the generated configuration immediately after installation:

cat .cursor/mcp.json

Expected structure:

{
  "cmd": "code-review-graph",
  "args": ["serve"],
  "cwd": "/path/to/your/project"
}

Claude Code MCP Configuration

Claude Code requires two linked configuration files:

1. .mcp.json (project root): A JSON-C file containing the project-scoped MCP settings, including repo_root, transport type, and server port. This file is shared across all platforms.
2. .claude/settings.json: A platform-specific settings file that points the Claude Code client to the .mcp.json configuration.

The dual-file structure allows Claude Code to resolve server paths while maintaining a central configuration that other editors can reuse.

Verification and Troubleshooting

After running the install command, confirm that the MCP server is correctly wired to your AI tool.

Verifying Cursor Integration

Cursor exposes MCP status through its command palette or CLI. Run the following to confirm the server is active:

cursor mcp status

You should see the code-review-graph server listed with a "connected" status and the correct repository path.

Verifying Claude Code Integration

Inside the Claude Code console, invoke the MCP diagnostic:

/mcp

This displays active MCP servers. Follow with /mcp help to list the available tools provided by code-review-graph, such as graph querying and review history retrieval.

Handling Path Changes and Windows Compatibility

If you relocate your repository or switch Python interpreters, you must re-run the install command to refresh the hard-coded paths inside .claude/settings.json. The tool does not dynamically resolve symbolic links or environment variables.

On Windows, ensure the generated .cursor/mcp.json does not use the cmd /c wrapper. According to the troubleshooting documentation (lines 68-78 in docs/TROUBLESHOOTING.md), you should specify the direct path to the Python executable or the installed CLI script to avoid shell escaping issues.

Summary

• Target specific platforms using code-review-graph install --platform <cursor|claude-code> to avoid generating unused configuration files.
• Cursor receives .cursor/mcp.json with stdio launch parameters.
• Claude Code receives .claude/settings.json referencing the shared .mcp.json project configuration.
• Re-run installation after moving the repository or changing interpreters to update absolute paths.
• Verify integration using cursor mcp status or the /mcp command in Claude Code.

Frequently Asked Questions

Can I install MCP configurations for both Cursor and Claude Code in the same repository?

Yes. Run the install command twice with different --platform flags, or omit the flag entirely to configure all detected platforms simultaneously. The shared .mcp.json ensures settings remain consistent across both tools.

Why does Claude Code use two configuration files while Cursor uses one?

Claude Code expects a separation between client settings and project-scoped server definitions. The .claude/settings.json file acts as a pointer to .mcp.json, allowing the latter to serve as a common configuration layer that other editors—including future platform support—can consume without modification.

What happens if I move my project directory after installation?

The generated files contain absolute paths to your repository root and Python interpreter. After moving the project, re-run code-review-graph install --platform <your-platform> to regenerate the configuration with updated paths. Failure to do so will result in "server not found" errors when the AI tool attempts to launch the MCP process.

Does the Windows configuration require special handling?

Yes. On Windows, avoid cmd /c wrappers in the .cursor/mcp.json command field. Specify the full path to the code-review-graph executable or the Python module directly to ensure proper stdio stream handling between the AI client and the server.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}