Delve MCP
# delve-mcp
MCP server for Delve debugger integration
This is a TypeScript-based MCP server that provides a complete interface to the Delve debugger for Go programs. It implements all major Delve commands and capabilities through MCP tools.
## Features
### Resources
- List and access debug sessions via `delve://` URIs
- Each session has metadata about its type, target, and port
- JSON representation for session details and state
### Tools
Debug, trace, and analyze Go programs with:
- Debug commands (`debug`, `attach`, `exec`, `test`)
- Core dump analysis
- Program tracing
- Replay debugging with `rr`
- DAP server support
- Breakpoint management with conditions
- Execution control (`continue`, `step`, `next`)
- Variable inspection and evaluation
- Backend selection (`native`, `lldb`, `rr`)
- Logging configuration
- Session management
### Environment Setup
The server requires:
- Go installed with Delve (`go install github.com/go-delve/delve/cmd/dlv@latest`)
- For replay functionality: Mozilla `rr` (https://github.com/mozilla/rr)
- Node.js and npm
## Installation
To install Delve MCP server:
```bash
npm install @dwisiswant0/delve-mcp
```
## Development
Install dependencies:
```bash
npm install
```
Build the server:
```bash
npm run build
```
For development with auto-rebuild:
```bash
npm run watch
```
### Configuration
To use with Claude Desktop, add the server config:
* On Linux: `~/.config/Claude/claude_desktop_config.json`.
* On MacOS: `~/Library/Application Support/Claude/claude_desktop_config.json`.
* On Windows: `%APPDATA%/Claude/claude_desktop_config.json`.
```json
{
"mcpServers": {
"delve-mcp": {
"command": "/path/to/delve-mcp/build/index.js"
}
}
}
```
## Available Tools
### Debug Commands
- **`debug`** - Debug a Go package in current directory or specified package
- **`attach`** - Attach to a running process by PID
- **`exec`** - Execute and debug a precompiled binary
- **`test`** - Debug tests in current package or specified package
- **`core`** - Examine a core dump file with associated executable
- **`dap`** - Start a Debug Adapter Protocol (DAP) server
- **`replay`** - Replay an rr trace recording
- **`trace`** - Trace program execution with function matching
### Control Commands
- **`setBreakpoint`** - Set a breakpoint with optional condition
- **`removeBreakpoint`** - Remove an existing breakpoint
- **`continue`** - Continue program execution
- **`next`** - Step over to next line
- **`step`** - Step into function call
- **`stepout`** - Step out of current function
- **`variables`** - List local variables in current scope
- **`evaluate`** - Evaluate expression in current scope
### Configuration Commands
- **`version`** - Get Delve version information
- **`setBackend`** - Configure debug backend (`native`, `lldb`, `rr`)
- **`configureLogging`** - Configure debug logging components
## Usage Examples
### Starting a Debug Session
```typescript
// Debug current package
{ name: "debug" }
// Debug with specific package and build flags
{
name: "debug",
arguments: {
package: "./cmd/myapp",
buildFlags: "-tags=integration"
}
}
```
### Managing Breakpoints
```typescript
// Set a breakpoint
{
name: "setBreakpoint",
arguments: {
sessionId: "abc123",
file: "main.go",
line: 42,
condition: "count > 5"
}
}
```
### Inspecting State
```typescript
// List variables
{
name: "variables",
arguments: {
sessionId: "abc123"
}
}
// Evaluate expression
{
name: "evaluate",
arguments: {
sessionId: "abc123",
expr: "myVar.Field"
}
}
```
## License
MIT.
TDQS
Scored across 19 tools
Tools have distinct purposes with clear descriptions. Some potential confusion exists between 'debug', 'exec', 'test', and 'attach' as different ways to start debugging, but descriptions differentiate them. Overall, agents can reliably distinguish tools.
All tools use a consistent single-word lowercase naming pattern (e.g., 'continue', 'step', 'setBreakpoint'). While not verb_noun, the pattern is uniform and predictable, making it easy for agents to understand.
19 tools is on the higher end for a debugger MCP server. While each tool serves a legitimate debugging function, the count exceeds the typical 3-15 range, making it somewhat heavy but still manageable.
Covers core debugging workflows like starting, stepping, breakpoints, and evaluation. However, missing common operations like listing breakpoints, viewing stack traces, and goroutine inspection indicate notable gaps for comprehensive debugging.