Overview
What LogTrace does
This desktop app records AI coding agent activity locally so you can review what happened after a session. It captures shell commands, file reads, notes, output, timestamps, project context, and risk labels in a local audit trail.
Windows is the current supported platform. Support for other desktop operating systems is planned for future releases.
It works with AI tools that can add a local HTTP MCP server, including Claude Code, Codex, Cursor, and other compatible MCP clients.
Logs stay on your computer unless you export or share them. The review flow is meant for developers who want to verify agent activity without sending their workflow to a hosted monitoring dashboard.
Installation
Install on Windows
Install LogTrace on the Windows computer where you use AI coding agents. The download page always points to the current Windows installer. Other desktop operating systems are not available yet, but they are planned.
Download v0.9.0 for WindowsConnecting agents
Connect your AI coding agent
Start LogTrace before connecting an agent. The app exposes a Streamable HTTP MCP server for Claude Code, Codex, Cursor, and other compatible MCP clients. If you change the MCP server port in LogTrace settings, replace 43117 with your configured port.
Automatic setup
The LogTrace app includes automatic setup for Codex and Claude Code in the connection settings. Instead of copying settings by hand, choose the target agent tab and run automatic setup.
- Codex
- Adds or updates the LogTrace MCP URL and logging rules in the global Codex config file and AGENTS.md.
- Claude Code
- Adds LogTrace as a user-scoped MCP server in Claude Code, then updates the global CLAUDE.md and related Claude hooks.
Manual setup
If you do not use automatic setup, configure these connection details in your MCP client.
http://127.0.0.1:43117/mcp- Server name
- LogTrace
- Transport
- Streamable HTTP
- Port
- 43117
Logging rules
Add instructions like the example below to the file used by your agent, such as AGENTS.md for Codex or CLAUDE.md for Claude Code. You can paste this as-is, or rewrite it to match your own project instructions. Tool approval behavior is controlled by each agent.
## LogTrace Logging Rules
- Use the LogTrace MCP tools as part of the work log when useful.
- Always use "Agent Name" for `agent_name`.
- Call `log_shell_execute` after running shell commands.
- Call `log_read_file` after reading important files.
- Call `log_note` for investigation notes and decisions.
- Write log entries in English.
- These logs are for debugging and local audit review."Agent Name" is a placeholder. Use "Codex" in AGENTS.md for Codex and "Claude" in CLAUDE.md for Claude Code.
-
log_shell_execute - Records shell commands and their output after execution.
-
log_read_file - Records important file reads with path and context.
-
log_note - Records investigation notes, decisions, and review context.
Reviewing logs
Review what happened
After an agent session, LogTrace shows commands, file reads, notes, outputs, and project context as separate entries. Project names are derived from the final folder name in the captured project root, so a path like C:\project\demo-app is shown as demo-app and can be used for filtering.
Timeline
Review agent activity in order, including timestamps and captured output.
Risk labels
Use risk labels to find entries that deserve closer attention.
Expanded details
Open an entry when you need command text, file paths, output, or notes.
Project names
Show project folder names derived from the captured project root.
Search and filtering
Narrow a session by risk, action, project, archived status, or matching text.
AI analysis
Use your own provider key
Available after license activation. AI analysis can use your own provider key for optional log review. You choose the provider and model, and keys are stored in OS secure storage when available.
Exchange Rate API is optional. If you add an Exchange Rate API key, LogTrace can display estimated API usage costs in your preferred currency.
| Provider | High Accuracy | Balanced | Low Cost |
|---|---|---|---|
| OpenAI | GPT-5.1 | GPT-5.4 mini | GPT-5.4 nano |
| Claude | Claude Sonnet 4.6 | Claude Haiku 4.5 | Not available |
| Gemini | Gemini 3.1 Pro Preview | Gemini 3 Flash Preview | Gemini 3.1 Flash-Lite Preview |
Troubleshooting
Fix common setup issues
LogTrace is not running
The desktop app is closed or did not finish starting. Open LogTrace again and wait until the app is ready before starting an agent session.
Agent cannot connect to the MCP endpoint
The agent is using the wrong endpoint or LogTrace is not listening locally. Review the server URL in LogTrace and confirm the agent uses the same MCP endpoint.
No new entries appear
The agent is not connected through LogTrace or the session started before connection. Reconnect the agent, start a new session, then run a simple command to confirm capture.
AI analysis is not shown
AI analysis is available after license activation and provider key setup. Activate your license, then add your provider key and model in the AI analysis settings.
AI analysis fails
The provider key, model name, or provider access may be invalid. Check the selected provider, confirm the model is available, and update the key if needed.
Installer download does not start
The browser blocked the redirect or the device is not a supported Windows computer. Open the download page on a Windows computer.
Support
Get help
If something does not work as expected, contact us with your LogTrace version, Windows version, agent name, and a short description of what happened. Improvement ideas and feature requests are welcome too.
Contact support