About Roo Activity Logger
Roo Activity Logger is an MCP server that automatically records activities of AI coding assistants such as Roo Code, Cline and Claude Code. Logs are saved as JSON files that can be searched and analyzed later, including for restoring context when resuming a task.
Key features
- 01Logs command executions, code generation and file operations
- 02Each log has level, summary, details and intention
- 03Supports parent, sequence and related ID grouping
- 04Searches logs by type, level, date and text
- 05Allows a separate save directory per activity
Use cases
- 01Restoring context and past intentions when resuming a task
- 02Reviewing operations performed by an AI coding assistant
- 03Analyzing development activity by date or type
MCP support
This project is an MCP server that exposes log_activity, get_log_files and search_logs tools to MCP clients.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMERoo Activity Logger
TL;DR
- What is it? — This is an MCP server that automatically logs AI coding assistant activities such as command executions and code generation (supports Roo Code, Cline, Claude Code, etc.).
- What does it do? — It saves activity history as JSON files, which you can search and analyze later.
- How do I use it? — Add it to your Claude Code, Cline, or Roo-Code settings to enable automatic activity logging.
Overview
Roo Activity Logger is an MCP (Model Context Protocol) server that automatically records AI coding assistant development activities — including command executions, code generation, file operations, and more. It supports Claude Code, Cline, Roo-Code, and other MCP-compatible AI assistants. All logs are saved in JSON format, making it easy to search, analyze, and restore context at any time.
How activity logging works
flowchart TD
A[Roo's Action] --> B[Select Activity Type]
B --> C[Log via log_activity]
C --> D[Provide Log Info<br>- Summary<br>- Details<br>- Intention<br>- Context]
D --> E[Specify Save Directory]
E --> F[Save as JSON file]
How to search logs and restore context when resuming a task
flowchart TD
G[Resume Task] --> H[Search with search_logs]
H --> I[Set Search Filters<br>- Type<br>- Date Range<br>- Text]
I --> J[Retrieve Related Activities<br>- Parent/Child<br>- Sequence<br>- Related IDs]
J --> K[Restore Context & Resume<br>- Review Past Intentions<br>- Understand Progress]
Sample Log Entry
Here is an example of a file_operation log entry saved as JSON:
{
"id": "75add15d-8d5b-4e60-b327-fde785050c86",
"timestamp": "2025-04-10T01:58:02.905Z",
"type": "file_operation",
"level": "info",
"summary": "Inserted mermaid diagram into README.md",
"details": {
"file": "README.md",
"operation": "insert_content",
"insertedLines": "mermaid code block",
"position": "after overview section"
},
"intention": "To visually explain the flow of saving and retrieving activities",
"context": "Improving documentation for Roo Activity Logger",
"parentId": "98280366-1de1-48e0-9914-b3a3409599b4"
}
Each log contains:
- Log level (
debug,info,warn,error) - Summary
- Details (any structured data)
- Intention / Purpose
- Context information
- Parent activity ID (for hierarchy)
- Sequence number (for ordering)
- Related activity IDs (for grouping)
Logs are:
- Saved as date-based JSON files
- Searchable by type, level, date, text, etc.
- Customizable — you can specify different save directories per activity
Features
-
Logs various activity types:
- Command executions (
command_execution) - Code generation (
code_generation) - File operations (
file_operation) - Errors (
error_encountered) - Decisions (
decision_made) - Conversations (
conversation)
- Command executions (
-
Each activity log includes:
- Unique ID
- Timestamp
- Activity type
- Summary, details, intention, context, and optional metadata
Usage (Recommended: via npx)
You can run Roo Activity Logger directly without cloning the repository by using npx.
Add this to your Cline, Roo-Code, or Claude Code configuration:
{
"mcpServers": {
"roo-activity-logger": {
"command": "npx",
"args": ["-y", "github:annenpolka/roo-logger"],
"env": {},
"disabled": false
}
}
}
Then, add prompts to your rule files (for Cline/Roo-Code) or CLAUDE.md (for Claude Code) to ensure logging, for example:
## Important
Always log activities using roo-activity-logger according to the logging rules.
## Preparation
Check the current context with `git status`.
Then, use roo-activity-logger's `search_logs` to review existing logs and identify current tasks.
Be sure to perform the logging steps.
## Logging
- Always use roo-activity-logger for all logs
- Include stack traces and execution context
- Record intention and context
For Developers: Local Setup
To develop or customize locally, clone the repository and build:
# Clone the repo
git clone https://github.com/annenpolka/roo-logger.git
cd roo-logger
# Install dependencies
npm install
# Build
npm run build
Example configuration to use your local build:
{
"mcpServers": {
"roo-activity-logger": {
"command": "node",
"args": ["/path/to/your/local/roo-logger/dist/index.js"], // adjust path accordingly
"env": {},
"disabled": false
}
}
}
Notes
- The specified directory will be created automatically if it does not exist.
MCP Tools
log_activity — Record an activity
A tool to record an activity.
Basic example
{
"type": "command_execution",
"summary": "Run npm command",
"intention": "Update project dependencies",
"context": "Preparing for new feature development",
"logsDir": "/absolute/path/to/logs/activity"
}
Parameters
| Name | Required | Type | Description |
|---|---|---|---|
type | Yes | string | Activity type (command_execution, code_generation, file_operation, error_encountered, decision_made, conversation) |
summary | Yes | string | Short summary of the activity |
intention | Yes | string | Purpose or intention |
context | Yes | string | Context information |
logsDir | Yes | string | Save directory (absolute path only) |
level | No | string | Log level (debug, info, warn, error). Default: info |
details | No | object | Additional details (any JSON) |
parentId | No | string | Parent activity ID |
sequence | No | number | Sequence number |
relatedIds | No | string[] | Related activity IDs |
Detailed example
{
"type": "file_operation",
"summary": "Update README file",
"intention": "Clarify documentation and improve usability",
"context": "Improvements based on user feedback",
"level": "info",
"details": {
"file": "README.md",
"operation": "update",
"changedLines": 15
},
"logsDir": "/absolute/path/to/logs/activity",
"sequence": 3,
"relatedIds": ["11223344-5566-7788-99aa-bbccddeeff00"]
}
get_log_files — List saved log files
Lists saved log files recursively. You can specify the maximum search depth.
Basic example
{
"logsDir": "/absolute/path/to/logs"
}
Parameters
| Name | Required | Type | Description |
|---|---|---|---|
logsDir | Yes | string | Directory to search (absolute path only) |
limit | No | number | Max files to retrieve (default: 10) |
offset | No | number | Number of files to skip (default: 0) |
logFilePrefix | No | string | Log file prefix (default: "roo-activity-") |
logFileExtension | No | string | Log file extension (default: ".json") |
maxDepth | No | number | Max directory depth (default: 3) |
search_logs — Search saved logs
Searches saved logs with various filters.
Basic example
{
"logsDir": "/absolute/path/to/logs"
}
{
"logsDir": "/absolute/path/to/logs",
"type": "command_execution"
}
Parameters
| Name | Required | Type | Description |
|---|---|---|---|
logsDir | Yes | string | Log directory (absolute path only) |
logFilePrefix | No | string | Log file prefix (default: "roo-activity-") |
logFileExtension | No | string | Log file extension (default: ".json") |
type | No | string | Filter by activity type (command_execution, code_generation, file_operation, error_encountered, decision_made, conversation) |
level | No | string | Filter by log level (debug, info, warn, error) |
startDate | No | string | Start date (YYYY-MM-DD) |
endDate | No | string | End date (YYYY-MM-DD) |
searchText | No | string | Search text in summary or details |
limit | No | number | Max logs to retrieve (default: 50) |
offset | No | number | Number of logs to skip (default: 0) |
parentId | No | string | Filter by parent activity ID |
sequenceFrom | No | number | Sequence number lower bound |
sequenceTo | No | number | Sequence number upper bound |
relatedId | No | string | Filter by related activity ID |
relatedIds | No | string[] | Filter by any of related activity IDs |
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.