csls-mcp

csls-mcp (C# Language Service - Model Context Protocol) is a long-running .NET 8 Console Application that provides a C# Language Service for AI agents. It leverages the Roslyn compiler platform to offer semantic understanding of C# code within a given workspace, integrating seamlessly with the Gemini CLI via the Model Context Protocol (MCP).

The idea behind this MCP server is to provide agents with more efficient tools to use instead of searchText or readFile or tempt them to read more code efficiently rather than guessing what the code does or where things are, so features will be added or changed based on agent usage experience.

🎯 Objective

The primary goal of csls-mcp is to enable AI agents to semantically interact with C# codebases. Instead of relying on textual search, agents can use csls-mcp to:

• Resolve C# symbols: Find classes, interfaces, methods, namespaces, properties, fields, etc.
• Find references: Locate where a symbol is used.
• List members of types: Enumerate methods, properties, and fields of a given class or interface.
• Obtain exact source code: Retrieve the precise source text of a symbol's declaration.

TODO:

• FindDeclaration: Locate where a symbol is declared.

This semantic understanding is crucial for tasks like intelligent code navigation, refactoring, bug fixing, and feature implementation by AI agents.

🧱 Architecture

• Type: .NET 8 Console Application
• Execution: Long-running process
• Communication: Model Context Protocol (MCP) over Standard I/O (STDIO)
• Workspace Management: Maintains the loaded C# solution/project in memory.
• Lifecycle: Automatically started and stopped by the Gemini CLI.
• Roslyn: Uses Microsoft.CodeAnalysis and MSBuildWorkspace for all semantic operations.

🚀 How to Compile

1. Navigate to the project directory:

   cd D:\CSLS-MCP\csls-mcp

2. Restore NuGet packages and build the project:

   dotnet build

   This will compile the application and place the executable in bin\Debug\net8.0\.

🤝 Integration with Gemini CLI

csls-mcp is designed to be integrated with Gemini CLI as an MCP server.

⚙️ Gemini CLI Configuration (.gemini/settings.json)

To enable Gemini CLI to use csls-mcp, add the following configuration to your .gemini/settings.json file in your project's root directory:

{
  "mcpServers": {
    "csharp": {
      "command": "D:\\CSLS-MCP\\csls-mcp\\bin\\Debug\\net8.0\\csls-mcp.exe",
      "args": ["--workspace", "${workspaceFolder}"],
      "trust": true
    }
  }
}

• csharp: This is the unique identifier (name) for this MCP server. You can choose any name you like.
• command: The absolute path to the compiled csls-mcp.exe executable. Make sure to update this path if your build output directory is different.
• args: Command-line arguments passed to the csls-mcp server.
  • --workspace ${workspaceFolder}: This tells the server the path to the current workspace (your C# solution/project). ${workspaceFolder} is a special variable that Gemini CLI resolves to the root of your current project.
• trust: Must be true to allow the server to execute and interact with your workspace.

✨ Example Usage from Gemini CLI

Once configured, Gemini CLI will automatically start csls-mcp when needed and allow the agent to use its exposed tools. Here are examples of how an AI agent might interact with the csharp MCP server:

1. getToolDeclarations (MCP Handshake)

This tool is used by the client (Gemini CLI) to discover the capabilities of the server.

Request:

{
  "id": "1",
  "tool": "getToolDeclarations",
  "input": {}
}

Response (example):

{
  "output": {
    "tools": [
      {
        "name": "resolveSymbol",
        "description": "Resolves a C# symbol (class, interface, method, etc.) and returns its kind, name, namespace, file, and line number.",
        "inputSchema": { "type": "object", "properties": { "symbol": { "type": "string" } }, "required": ["symbol"] }
      },
      {
        "name": "getSymbolSource",
        "description": "Retrieves the exact source code of a given C# symbol.",
        "inputSchema": { "type": "object", "properties": { "symbol": { "type": "string" } }, "required": ["symbol"] }
      },
      {
        "name": "findReferences",
        "description": "Finds all references to a given C# symbol within the loaded workspace.",
        "inputSchema": { "type": "object", "properties": { "symbol": { "type": "string" } }, "required": ["symbol"] }
      },
      {
        "name": "listMembers",
        "description": "Lists methods, properties, and fields of a specified C# type.",
        "inputSchema": { "type": "object", "properties": { "symbol": { "type": "string" } }, "required": ["symbol"] }
      }
    ]
  },
  "id": "1"
}

2. resolveSymbol

Request:

{
  "id": "2",
  "tool": "resolveSymbol",
  "input": {
    "symbol": "MyTestClass"
  }
}

Response (example, assuming MyTestClass is in dummy-project/Program.cs):

{
  "output": {
    "kind": "Class",
    "name": "MyTestClass",
    "namespace": "dummy_project",
    "file": "D:\\CSLS-MCP\\dummy-project\\Program.cs",
    "line": 5
  },
  "id": "2"
}

3. getSymbolSource

Request:

{
  "id": "3",
  "tool": "getSymbolSource",
  "input": {
    "symbol": "MyTestMethod"
  }
}

Response (example, showing source of MyTestMethod):

{
  "output": {
    "file": "D:\\CSLS-MCP\\dummy-project\\Program.cs",
    "source": "        public void MyTestMethod(string param)\n        {\n            Console.WriteLine($\"Parameter: {param}, Property: {MyProperty}, Field: {_myField}\");\n        }"
  },
  "id": "3"
}

4. findReferences

Request:

{
  "id": "4",
  "tool": "findReferences",
  "input": {
    "symbol": "MyTestClass"
  }
}

Response (example, showing a reference in Program.cs):

{
  "output": [
    {
      "file": "D:\\CSLS-MCP\\dummy-project\\Program.cs",
      "line": 27
    }
  ],
  "id": "4"
}

5. listMembers

Request:

{
  "id": "5",
  "tool": "listMembers",
  "input": {
    "symbol": "MyTestClass"
  }
}

Response (example):

{
  "output": {
    "methods": [
      "MyTestMethod(String param)"
    ],
    "properties": [
      "String MyProperty"
    ],
    "fields": [
      "Int32 _myField"
    ]
  },
  "id": "5"
}