Log MCP Server

# Changelog All notable changes to this project will be documented in this file. ## [0.4.6] - 2025-12-11 ### Changed - **Improved `list_log_files` description**: Now triggers when user says 'inspect', 'inspector', or 'logs' - Better discoverability as entry point for log inspection workflows - Clearer guidance that this tool should be used first before other inspection tools ## [0.4.5] - 2025-12-10 ### Fixed - **Accurate `max_tokens` enforcement**: Token counting now includes all output overhead - Header/metadata text is now counted in the token budget - Line number prefixes (`{i:6d} | `) are now counted per line (~3 tokens each) - Footer text is now reserved in the token budget - Previously, requesting 15000 tokens could return ~16000+ tokens due to uncounted overhead - Affected tools: `read_log_paginated`, `search_log_file`, `head_log`, `tail_log`, `read_log_range`, `find_errors` ## [0.4.4] - 2025-12-08 ### Added - **`find_errors` tool**: Quickly finds error lines by matching common error patterns - Matches: ERROR, FATAL, CRITICAL, Exception, Traceback, Failed, panic, etc. - `context_lines` parameter: lines to show before/after each error (default: 2) - `include_warnings` parameter: also match WARNING patterns (default: false) - `max_tokens` parameter: limit by token count (default: 4000) - Shows line numbers with `>>>` markers for error lines ### Rationale - AI assistants can quickly diagnose issues without reading entire log files - Common patterns cover most programming languages and frameworks - Token-based pagination ensures manageable output size ## [0.4.2] - 2025-12-08 ### Added - **`head_log` tool**: Reads the beginning of a log file (like Unix 'head' command) with token-based pagination - `lines` parameter: specify exact number of lines to read - `max_tokens` parameter: limit by token count (default: 4000) - **`tail_log` tool**: Reads the end of a log file (like Unix 'tail' command) with token-based pagination - `lines` parameter: specify exact number of lines to read - `max_tokens` parameter: limit by token count (default: 4000) - **`read_log_range` tool**: Reads a specific range of lines from a log file - `start_line` and `end_line` parameters for precise range selection - `max_tokens` parameter: limit by token count (default: 4000) - All tools show line numbers and file statistics ### Rationale - Quick access to start or end of log files without full pagination setup - `tail_log` ideal for checking recent log entries during debugging - `read_log_range` enables reading arbitrary sections of log files - `lines` parameter provides Unix-like head/tail behavior - Consistent token-based approach across all tools ## [0.4.1] - 2025-12-08 ### Added - **Token-based pagination for `get_log_content`**: Now uses `max_tokens` (default: 4000, max: 100000) to limit output - Automatic truncation with guidance to use `read_log_paginated` for large files - File size and token count shown in output header ### Rationale - Previously `get_log_content` returned entire file without limits - Large files could overwhelm AI context windows - Consistent token-based approach across all tools ## [0.4.0] - 2025-12-07 ### Changed - **Token-based pagination for `search_log_file`**: Now uses `max_tokens` (default: 4000) instead of `max_matches` for pagination - `max_matches` parameter deprecated but still supported for backward compatibility - Search results now show mode (token-based vs match-based) and estimated token count - Improved pagination to respect AI context limits by estimating tokens for each match with its context ### Rationale - Search results can be large when matches include context lines - Token-based pagination provides more predictable memory usage for AI assistants - Consistent with `read_log_paginated` pagination strategy - Better user experience by avoiding context limit issues ## [0.2.5] - 2025-12-07 ### Added - File modification tracking for `read_log_paginated` - `expected_size` and `expected_mtime` parameters to detect file changes during pagination - Automatic warnings when log files are modified between pagination calls - File metadata (size, mtime) included in all responses - Helpful "For next call, use:" message with correct parameters ### Fixed - Race condition when reading actively-written log files - Inconsistent line numbers when files are modified during pagination ### Rationale - Log files are often actively written to during debugging - File changes between pagination calls cause line number shifts - Detecting changes allows AI to restart from beginning or warn user - Provides visibility into file state for better debugging ## [0.2.4] - 2025-12-07 ### Changed - **BREAKING (but backward compatible)**: `read_log_paginated` now uses token-based pagination instead of line-based - Default pagination is now `max_tokens=4000` instead of `num_lines=100` - Uses ~4 characters per token estimation for consistent context usage - `num_lines` parameter deprecated but still supported for backward compatibility - Output now shows mode (token-based vs line-based) and estimated token count ### Rationale - Log lines vary drastically in length (10-10000+ characters) - Token limits matter for AI context windows, not line counts - Provides more predictable and useful pagination for AI assistants ## [0.2.3] - 2025-12-07 ### Added - `context_before` and `context_after` parameters for `search_log_file` tool - Separate control of context lines before and after matches (like grep's -B and -A options) - Support for asymmetric context display (e.g., 5 lines before, 2 lines after) ### Changed - `context_lines` now serves as default for both before and after context - Context display in search results shows separate before/after counts when different ## [0.2.2] - 2025-12-07 ### Changed - Package renamed to log-inspector-mcp for PyPI compatibility - Updated GitHub repository URL in package metadata - Documentation improvements for configuration examples ## [0.2.1] - 2025-12-07 ### Changed - Updated package metadata with author information - Minor documentation improvements ## [0.2.0] - 2025-12-07 ### Added - `read_log_paginated` tool for reading large log files in chunks with line numbers - `search_log_file` tool for regex-based searching with: - Configurable context lines (like grep -C) - Case-sensitive/insensitive search - Pagination support for search results - Visual markers (>>>) for matching lines - Support for handling large log files efficiently - **Multiple log directory support:** - `--log-dir` command-line argument (can be specified multiple times) - `LOG_MCP_DIR` environment variable (colon-separated paths) - `list_log_files` scans all configured directories - Auto-resolves filenames across all directories - Configurable log directory with priority: CLI args > env var > default ### Changed - Updated `runtime-logs` prompt to mention all 4 available tools and all directories - Enhanced tool descriptions to guide when to use each tool - Improved error messages when log directories are not accessible - `list_log_files` now shows which directories are being scanned - **Comprehensive README documentation:** - Added detailed "What is this?" section - Included use cases and workflow examples - Added application logging configuration for Node.js, Python, Java, Rust, and Go - Added IDE configuration examples (VSCode, IntelliJ/PyCharm) - Added example end-to-end debugging workflow ## [0.1.0] - 2025-12-07 ### Added - Initial release of log-mcp MCP server - `list_log_files` tool to list all log files in $XDG_RUNTIME_DIR/log - `get_log_content` tool to read specific log files - `runtime-logs` prompt that automatically informs AI about log inspection capabilities - Security checks to ensure file access is restricted to the log directory - Proper pip-installable package structure with setuptools - MIT License - Comprehensive README with installation and usage instructions - AI guidance system that proactively suggests log inspection when users report errors