Debugger MCP Server

A JetBrains IDE plugin that exposes an MCP (Model Context Protocol) server, giving AI coding assistants full programmatic control over the debugger. Set breakpoints, step through code, inspect variables, and evaluate expressions—all driven autonomously by your AI assistant.

Verified in CI (IntelliJ Plugin Verifier): IntelliJ IDEA, PyCharm, WebStorm, GoLand. Community-tested: RustRover, Android Studio, PhpStorm May work (untested): RubyMine, CLion, DataGrip

Debugger MCP Server gives AI coding assistants complete control over the IDE's debugger through the Model Context Protocol (MCP). Let your AI assistant debug code autonomously—from setting breakpoints to inspecting variables to stepping through execution.

Features

Debug Session Management

• Start/Stop Sessions - Launch any run configuration in debug mode
• Rich Status - Get comprehensive state in a single call (variables, stack, source context)
• Multi-Session Support - Manage multiple concurrent debug sessions

Breakpoint Management

• Line Breakpoints - Set breakpoints at any valid location
• Conditional Breakpoints - Add conditions that must be true to pause
• Tracepoints - Log messages without pausing execution

Execution Control

• Step Over/Into/Out - Navigate through code line by line
• Resume & Pause - Control execution flow
• Run to Line - Continue execution until a specific line

Variable Inspection

• View Variables - Inspect local variables, arguments, and object fields
• Modify Values - Change variable values during debugging

Expression Evaluation

• Evaluate Expressions - Run arbitrary expressions in the current context
• Code Fragments - Execute multi-line code snippets
• Safety Controls - Choose unrestricted evaluation, a default risky-operation blocklist, or conservative read-only checks with optional custom regex block rules

Stack & Thread Navigation

• Stack Traces - View the complete call stack with source locations
• Frame Selection - Switch context to any stack frame
• Thread Listing - See all threads and their states

Companion Skill

• AI Debugging Guidance - Bundled companion skill teaches AI agents optimal debugger tool usage
• One-Click Install - Install to .claude/skills/ or export as .skill/.zip

Why Use This Plugin?

Unlike manual debugging, this plugin enables:

• Autonomous AI Debugging - Your AI assistant can debug code without human intervention
• Rich Context in Single Calls - Get variables, stack, and source in one request
• Programmatic Breakpoint Control - Set conditional breakpoints with complex expressions
• Cross-IDE Compatibility - Works with any JetBrains IDE that supports XDebugger
• 23 Comprehensive Tools - Full debugging capability through MCP
• Configurable Server - IDE-specific ports with customizable host binding

Perfect for AI-assisted development workflows where you want your assistant to investigate bugs, validate fixes, or explore code behavior autonomously.

Table of Contents

• Installation
• Quick Start
• Client Configuration
• Available Tools
• Multi-Project Support
• Tool Window
• Error Codes
• Settings
• Requirements
• Contributing

Installation

Using the IDE built-in plugin system

Settings/Preferences > Plugins > Marketplace > Search for "Debugger MCP Server" > Install

Using JetBrains Marketplace

Go to JetBrains Marketplace and install it by clicking the Install to ... button.

Manual Installation

Download the latest version from the JetBrains Marketplace versions page and install it manually: Settings/Preferences > Plugins > ⚙️ > Install plugin from disk...

Quick Start

1. Install the plugin and restart your JetBrains IDE
2. Open a project - the MCP server starts automatically on an IDE-specific port
3. Find your IDE port: Settings > Tools > Debugger MCP Server (each IDE has a unique default port, e.g., 29190 for IntelliJ IDEA)
4. Configure your AI assistant with the server URL: http://127.0.0.1:{PORT}/debugger-mcp/streamable-http
5. Use the tool window (bottom panel: "Debugger MCP Server") to copy configuration or monitor commands

Using the "Install on Coding Agents" Button

The easiest way to configure your AI assistant:

1. Open the "Debugger MCP Server" tool window (bottom panel)
2. Click the prominent "Install on Coding Agents" button on the right side of the toolbar
3. A popup appears with three sections:
   • Install Now - For Claude Code CLI and Codex CLI: Runs the installation command automatically
   • Copy Configuration - For other clients (Gemini CLI, Cursor, etc.): Copies the JSON config to your clipboard
   • Generic MCP Config - Streamable HTTP or Legacy SSE configuration for any MCP client
4. For "Copy Configuration" clients, paste the config into the appropriate config file

Example Workflow

Just tell your AI assistant:

"Debug the calculateTotal function—set a breakpoint at line 42, run the tests in debug mode, and show me the variable values when it pauses."

Or for more complex debugging:

"There's a bug in UserService. Set a breakpoint at line 42, run the tests in debug mode, and when it breaks, show me the stack trace and all local variables."

Client Configuration

Claude Code (CLI)

The easiest way is to use the "Install on Coding Agents" button in the IDE's tool window—it generates the correct command with your IDE-specific server name and port.

Or run this command manually in your terminal (replace <ide>-debugger and port with your IDE's values):

claude mcp add --transport http intellij-debugger http://127.0.0.1:29190/debugger-mcp/streamable-http --scope user

IDE-specific server names and default ports:

Options:

• --scope user - Adds globally for all projects
• --scope project - Adds to current project only

To remove: claude mcp remove intellij-debugger (use your IDE's name)

Codex CLI

The easiest way is to use the "Install on Coding Agents" button in the IDE's tool window—it generates the correct command with your IDE-specific server name and port.

Or run this command manually in your terminal (replace <ide>-debugger and port with your IDE's values):

codex mcp add intellij-debugger --url http://127.0.0.1:29190/debugger-mcp/streamable-http

To remove: codex mcp remove intellij-debugger (use your IDE's name)

Gemini CLI

Add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "intellij-debugger": {
      "httpUrl": "http://127.0.0.1:29190/debugger-mcp/streamable-http"
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project root or ~/.cursor/mcp.json globally:

{
  "mcpServers": {
    "intellij-debugger": {
      "url": "http://127.0.0.1:29190/debugger-mcp/streamable-http"
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "intellij-debugger": {
      "serverUrl": "http://127.0.0.1:29190/debugger-mcp/streamable-http"
    }
  }
}

VS Code (Generic MCP)

{
  "mcp.servers": {
    "intellij-debugger": {
      "url": "http://127.0.0.1:29190/debugger-mcp/streamable-http"
    }
  }
}

Note: Replace intellij-debugger and the port with your IDE's server name and default port (see table above).

Note: The port can be changed in Settings > Tools > Debugger MCP Server.

Available Tools

The plugin provides 23 MCP tools organized by category:

Run Configuration Tools

Debug Session Tools

Breakpoint Tools

Execution Control Tools

Stack & Thread Tools

Variable Tools

Navigation Tools

Evaluation Tools

Note: For detailed tool documentation with parameters, examples, and response formats, see USAGE.md.

Multi-Project Support

When multiple projects are open in a single IDE window, you must specify which project to use with the project_path parameter:

{
  "name": "set_breakpoint",
  "arguments": {
    "project_path": "/Users/dev/myproject",
    "file_path": "/Users/dev/myproject/src/Main.java",
    "line": 42
  }
}

If project_path is omitted:

• Single project open: That project is used automatically
• Multiple projects open: An error is returned with the list of available projects

Tool Window

The plugin adds a "Debugger MCP Server" tool window (bottom panel) that shows:

• Server Status: Running indicator with server URL and port
• Agent Rule Tip: Copy a rule for your AI agent's config to prefer debugger MCP tools
• Project Name: Currently active project
• Command History: Log of all MCP tool calls with:
  • Timestamp
  • Tool name
  • Status (Success/Error/Pending)
  • Parameters and results (expandable)
  • Execution duration
• Filters: Filter history by tool name, status, or search text

Tool Window Actions

Error Codes

JSON-RPC Standard Errors

Tool Failures

Tool failures are not JSON-RPC errors. A tool that fails returns a successful result whose payload carries isError: true and a human-readable message in content[0].text, so the model can read what went wrong and retry:

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [{ "type": "text", "text": "Session must be paused to get variables" }],
    "isError": true
  }
}

The JSON-RPC error channel is reserved for protocol-level problems — malformed JSON, an unparseable message, an unknown method (not an unknown tool).

Project resolution is the one failure that returns structured JSON in content[0].text, because an agent that picked the wrong project needs the list of real ones to retry with:

{ "error": "multiple_projects_open", "message": "...", "available_projects": [ ... ] }

error is one of no_project_open, project_not_found or multiple_projects_open.

Settings

Configure the plugin at Settings > Tools > Debugger MCP Server:

Evaluate Expression Safety

evaluate_expression is powerful because IDE debuggers can call methods and run code in the debugged process. The safety setting is a best-effort guardrail, not a sandbox.

• Unrestricted - No plugin-side filtering. Expressions may call methods, mutate state, access files/network, or run any operation accepted by the debugger.
• Default blocklist - Blocks common high-risk categories: process execution, JVM termination, filesystem access, network access, reflection/access bypass, native code loading, and environment/system property access. Custom regex rules also apply.
• Read-only - Includes the default blocklist and custom regex rules, then rejects assignments, increment/decrement operations, code fragments, constructors, and method calls that cannot be proven read-only.

Custom regex rules are additional blocked patterns. They do not replace the built-in restrictions.

Requirements

• JetBrains IDE 2025.2 or later (any IDE based on IntelliJ Platform)
• JVM 21 or later
• MCP Protocol negotiated per client, from 2024-11-05 up to 2025-11-25, over the official MCP Kotlin SDK

Supported IDEs

Verified in CI (IntelliJ Plugin Verifier): IntelliJ IDEA, PyCharm, WebStorm, GoLand. Community-tested: RustRover, Android Studio, PhpStorm May work (untested): RubyMine, CLion, DataGrip, Aqua, DataSpell, Rider

Architecture

The plugin runs an embedded Ktor CIO server on an IDE-specific port and supports three MCP transports. The protocol itself — JSON-RPC framing, the initialize handshake, version negotiation, session lifecycle and batching — is handled by the official MCP Kotlin SDK; the plugin owns only the routes, the loopback-Origin guard, and the debugger tools themselves.

Streamable HTTP Transport (Primary)

AI Assistant ──────► POST /debugger-mcp/streamable-http   (JSON-RPC with Mcp-Session-Id header)
                     ◄── JSON-RPC response                (immediate HTTP response)
             ──────► GET /debugger-mcp/streamable-http    (server → client event stream)
                     ◄── event: message                   (notifications, progress)
             ──────► DELETE /debugger-mcp/streamable-http  (session termination)

Legacy HTTP+SSE Transport

AI Assistant ──────► GET /debugger-mcp/sse           (establish SSE stream)
                     ◄── event: endpoint             (receive POST URL with sessionId)
             ──────► POST /debugger-mcp?sessionId=x  (JSON-RPC requests)
                     ◄── HTTP 202 Accepted
                     ◄── event: message              (JSON-RPC response via SSE)

Stateless HTTP (Convenience)

AI Assistant ──────► POST /debugger-mcp              (JSON-RPC requests, no session)
                     ◄── JSON-RPC response           (immediate HTTP response)

This approach:

• Modern clients - Streamable HTTP with session management and a server → client event stream
• Legacy clients - Full HTTP+SSE transport for pre-2025 clients
• Protocol version is negotiated per client, from 2024-11-05 up to 2025-11-25
• Simple clients - Stateless HTTP for quick request/response without sessions
• Each IDE gets a unique default port to avoid conflicts when multiple IDEs run simultaneously
• Works with any MCP-compatible client
• Supports CORS for browser-based clients

Contributing

Contributions are welcome! See CONTRIBUTING.md for the build/test quickstart and the project-specific tripwires (JDK 21, pumpingEdt, golden-contract regeneration). For security concerns, see SECURITY.md. In short:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Run tests: ./gradlew test
5. Submit a pull request

Development Setup

# Build the plugin
./gradlew build

# Run IDE with plugin installed
./gradlew runIde

# Run tests
./gradlew test

# Run plugin verification
./gradlew runPluginVerifier

License

This project is licensed under the MIT License - see the LICENSE file for details.

Plugin based on the IntelliJ Platform Plugin Template.