gdb-cli
A GDB debugging tool designed for AI Agents (Claude Code, etc.)
README
GDB CLI for AI
A GDB debugging tool designed for AI Agents (Claude Code, etc.). Uses a "thin client CLI + GDB built-in Python RPC Server" architecture, enabling stateful GDB debugging through Bash.
Features
- Core Dump Analysis: Load core dumps with symbols resident in memory for millisecond-level response
- Live Attach Debugging: Attach to running processes with non-stop mode support
- Structured JSON Output: All commands output JSON with automatic truncation/pagination and operation hints
- Security Mechanisms: Command whitelist, heartbeat timeout auto-cleanup, idempotency guarantees
- Database-Optimized: scheduler-locking, large object pagination, multi-thread truncation
Requirements
- Python: 3.6.8+
- GDB: 9.0+ with Python support enabled
- OS: Linux
Check GDB Python Support
# Check if GDB has Python support
gdb -nx -q -batch -ex "python print('OK')"
# If system GDB lacks Python, check GCC Toolset (RHEL/CentOS)
/opt/rh/gcc-toolset-13/root/usr/bin/gdb -nx -q -batch -ex "python print('OK')"
Installation
# Install from PyPI
pip install gdb-cli
# Or install from GitHub
pip install git+https://github.com/Cerdore/gdb-cli.git
# Or clone and install locally
git clone https://github.com/Cerdore/gdb-cli.git
cd gdb-cli
pip install -e .
Environment check
gdb-cli env-check
## Quick Start
### 1. Load Core Dump
```bash
gdb-cli load --binary ./my_program --core ./core.12345
Output:
{
"session_id": "f465d650",
"mode": "core",
"binary": "./my_program",
"core": "./core.12345",
"gdb_pid": 12345,
"status": "started"
}
If your system's default GDB doesn't have Python support, specify it with
--gdb-path:gdb-cli load --binary ./my_program --core ./core.12345 \ --gdb-path /opt/rh/gcc-toolset-13/root/usr/bin/gdb
2. Debugging Operations
All operations use --session / -s to specify the session ID:
SESSION="f465d650"
# List threads
gdb-cli threads -s $SESSION
# Get backtrace (default: current thread)
gdb-cli bt -s $SESSION
# Get backtrace for a specific thread
gdb-cli bt -s $SESSION --thread 3
# Evaluate C/C++ expressions
gdb-cli eval-cmd -s $SESSION "my_struct->field"
# Access array elements
gdb-cli eval-element -s $SESSION "my_array" --index 5
# View local variables
gdb-cli locals-cmd -s $SESSION
# Execute raw GDB commands
gdb-cli exec -s $SESSION "info registers"
# Check session status
gdb-cli status -s $SESSION
3. Session Management
# List all active sessions
gdb-cli sessions
# Stop a session
gdb-cli stop -s $SESSION
4. Live Attach Debugging
# Attach to a running process (default: scheduler-locking + non-stop)
gdb-cli attach --pid 9876
# Attach with symbol file
gdb-cli attach --pid 9876 --binary ./my_program
# Allow memory modification and function calls
gdb-cli attach --pid 9876 --allow-write --allow-call
Full Command Reference
load — Load Core Dump
gdb-cli load --binary <path> --core <path> [options]
--binary, -b Executable file path (required)
--core, -c Core dump file path (required)
--sysroot sysroot path (for cross-machine debugging)
--solib-prefix Shared library prefix
--source-dir Source code directory
--timeout Heartbeat timeout in seconds (default: 600)
--gdb-path GDB executable path (default: "gdb")
attach — Attach to Process
gdb-cli attach --pid <pid> [options]
--pid, -p Process PID (required)
--binary Executable file path (optional)
--scheduler-locking Enable scheduler-locking (default: true)
--non-stop Enable non-stop mode (default: true)
--timeout Heartbeat timeout in seconds (default: 600)
--allow-write Allow memory modification
--allow-call Allow function calls
threads — List Threads
gdb-cli threads -s <session> [options]
--range Thread range, e.g., "3-10"
--limit Maximum return count (default: 20)
--filter-state Filter by state ("running" / "stopped")
bt — Backtrace
gdb-cli bt -s <session> [options]
--thread, -t Specify thread ID
--limit Maximum frame count (default: 30)
--full Include local variables
--range Frame range, e.g., "5-15"
eval-cmd — Evaluate Expression
gdb-cli eval-cmd -s <session> <expr> [options]
--max-depth Recursion depth limit (default: 3)
--max-elements Array element limit (default: 50)
eval-element — Access Array/Container Elements
gdb-cli eval-element -s <session> <expr> --index <N>
exec — Execute Raw GDB Command
gdb-cli exec -s <session> <command>
--safety-level Safety level (readonly / readwrite / full)
thread-apply — Batch Thread Operations
gdb-cli thread-apply -s <session> <command> --all
gdb-cli thread-apply -s <session> <command> --threads "1,3,5"
Output Examples
threads
{
"threads": [
{"id": 1, "global_id": 1, "state": "stopped"},
{"id": 2, "global_id": 2, "state": "stopped"}
],
"total_count": 5,
"truncated": true,
"current_thread": {"id": 1, "global_id": 1, "state": "stopped"},
"hint": "use 'threads --range START-END' for specific threads"
}
eval-cmd
{
"expression": "(int)5+3",
"value": 8,
"type": "int",
"size": 4
}
bt
{
"frames": [
{"number": 0, "function": "crash_thread", "address": "0x400a1c", "file": "test.c", "line": 42},
{"number": 1, "function": "start_thread", "address": "0x7f3fa2e13fa"}
],
"total_count": 2,
"truncated": false
}
Security Mechanisms
Command Whitelist (Attach Mode)
| Safety Level | Allowed Commands |
|---|---|
readonly (default) |
bt, info, print, threads, locals, frame |
readwrite |
+ set variable |
full |
+ call, continue, step, next |
quit, kill, shell, signal are always blocked.
Heartbeat Timeout
Automatically detaches and quits after 10 minutes of inactivity by default. Configurable via --timeout.
Idempotency
Only one session per PID / Core file is allowed. Repeated load/attach returns the existing session_id.
Cross-Machine Core Dump Debugging
When analyzing core dumps from other machines, shared library paths may differ:
# Set sysroot (path prefix replacement)
gdb-cli load --binary ./my_program --core ./core.1234 \
--sysroot /path/to/target/rootfs
# Set source directory (for source-level debugging)
gdb-cli load --binary ./my_program --core ./core.1234 \
--source-dir /path/to/source
Development
Project Structure
src/gdb_cli/
├── cli.py # CLI entry point (Click)
├── client.py # Unix Socket client
├── launcher.py # GDB process launcher
├── session.py # Session metadata management
├── safety.py # Command whitelist filter
├── formatters.py # JSON output formatting
├── env_check.py # Environment check
├── errors.py # Error classification
└── gdb_server/
├── gdb_rpc_server.py # RPC Server core
├── handlers.py # Command handlers
├── value_formatter.py # gdb.Value serialization
└── heartbeat.py # Heartbeat timeout management
skills/
└── gdb-cli/ # Claude Code skill for intelligent debugging
├── SKILL.md # Skill definition
└── evals/ # Test cases for skill evaluation
Run Tests
pip install -e ".[dev]"
pytest tests/ -v
End-to-End Testing
Requires GDB with Python support. Use the crash test program in tests/crash_test/:
# Compile test program
cd tests/crash_test
gcc -g -pthread -o crash_test crash_test_c.c
# Generate coredump
ulimit -c unlimited
./crash_test # Will SIGSEGV
# Find core file
ls /path/to/core_dumps/core-crash_test-*
# Run E2E test
gdb-cli load --binary ./crash_test --core /path/to/core \
--gdb-path /opt/rh/gcc-toolset-13/root/usr/bin/gdb
Known Limitations
- No
target remotesupport (use SSH for remote debugging, see below) - No multi-inferior debugging support
- GDB 12.x Guile pretty printers are not thread-safe, workaround via
format_string(raw=True) - GDB embedded Python version may be older (e.g., 3.6.8), code has compatibility handling
Remote Debugging via SSH
Install and run on remote machine in one command:
ssh user@remote-host "pip install git+https://github.com/Cerdore/gdb-cli.git && gdb-cli load --binary ./my_program --core ./core.12345"
Or install first, then debug:
# Install on remote
ssh user@remote-host "pip install git+https://github.com/Cerdore/gdb-cli.git"
# Run debugging
ssh user@remote-host "gdb-cli load --binary ./my_program --core ./core.12345"
Claude Code Skills
This project includes a gdb-cli skill for Claude Code that provides intelligent debugging assistance by combining source code analysis with runtime state inspection.
Install the Skill
bunx skills add https://github.com/Cerdore/gdb-cli --skill=gdb-cli
Usage in Claude Code
/gdb-cli
# Or describe your debugging need:
I have a core dump at ./core.1234 and binary at ./myapp. Help me debug it.
Features
- Source Code Correlation: Automatically reads source files around crash points
- Deadlock Detection: Identifies circular wait patterns in multi-threaded programs
- Safety Warnings: Alerts about production environment risks when attaching to live processes
- Structured Reports: Generates analysis with root cause hypotheses and next steps
See skills/README.md for more details.
License
Apache License 2.0
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.