Forensic Log MCP Server

A high-performance Model Context Protocol (MCP) server for log analysis, powered by Polars and custom SIMD-accelerated parsers.

Give Claude the ability to analyze massive log files (gigabytes of data) that would never fit in its context window.

Features

Fast Aggregations: 5-50x faster than awk on GROUP BY operations
SIMD-Accelerated: Custom parsers using memchr for structured queries
Multi-Format: Apache, Nginx, Syslog, JSON Lines, CSV/TSV
Streaming: Handles files larger than RAM via lazy evaluation
AI-Native: Designed for Claude integration via MCP protocol

Performance

MCP excels at aggregation queries on large files (1M lines, 125MB Apache log):

Operation	awk	MCP	Speedup
Group by IP	0.26s	0.048s	5x faster
Group by method	1.31s	0.047s	28x faster
Group by user_agent	2.40s	0.049s	50x faster
Group by referer	1.14s	0.046s	25x faster
Sum size	0.37s	0.047s	8x faster
JSON group by (vs jq)	1.65s	0.16s	11x faster

Honest Assessment:

MCP dominates GROUP BY/aggregation queries (5-50x faster)
grep is ~24x faster for simple line counting (grep -c)
MCP has ~57ms Python/MCP overhead, negligible on large files

See benchmark/BENCHMARK.md for detailed methodology and honest comparison.

Quick Start

Installation

# Clone the repository
git clone https://github.com/TLinvest/forensic-log-mcp.git
cd forensic-log-mcp

# Build the MCP server
cd mcp
cargo build --release

The binary will be at mcp/target/release/forensic-log-mcp.

Configuration

Add to your Claude Code project's .mcp.json:

{
  "mcpServers": {
    "forensic-logs": {
      "type": "stdio",
      "command": "/path/to/forensic-log-mcp",
      "args": []
    }
  }
}

Or add globally to ~/.claude.json.

Usage Examples

Once configured, Claude can analyze your logs directly:

"Find all 500 errors in my nginx access log"
→ Uses analyze_logs with filter_status="500"

"Count requests by IP address"
→ Uses aggregate_logs with group_by="ip"

"Search for timeout errors in the last hour"
→ Uses search_pattern with regex matching

"Show me the error rate by hour"
→ Uses time_analysis with bucket="hour"

Available Tools

Tool	Description
`get_log_schema`	Discover columns and sample data from a log file
`analyze_logs`	Filter, group, and sort log data
`aggregate_logs`	Perform statistical aggregations (count, sum, avg, min, max)
`search_pattern`	Search for regex patterns
`time_analysis`	Analyze logs over time with bucketing

Supported Formats

Format	Auto-Detected	SIMD Fast Path
Apache/Nginx Combined	Yes	Yes
Syslog (RFC 3164/5424)	Yes	Yes
JSON Lines (NDJSON)	Yes	Polars-native
CSV/TSV	Yes	Polars-native

Project Structure

forensic-log-mcp/
├── mcp/                    # MCP server source code
│   ├── src/
│   │   ├── main.rs        # Entry point
│   │   ├── tools/         # MCP tool implementations
│   │   ├── parsers/       # Log format parsers
│   │   │   ├── apache_simd.rs   # SIMD Apache parser
│   │   │   ├── syslog_simd.rs   # SIMD Syslog parser
│   │   │   └── ...
│   │   └── engine/        # Query engine
│   ├── Cargo.toml
│   └── README.md          # Detailed MCP documentation
├── benchmark/             # Performance benchmarks
│   ├── BENCHMARK.md       # Detailed results
│   ├── run_benchmark.sh   # Benchmark runner
│   └── scripts/           # Benchmark utilities
├── LICENSE
├── CONTRIBUTING.md
└── README.md              # This file

How It's So Fast

SIMD-Accelerated Parsing: Uses memchr for vectorized field detection
Zero-Copy Processing: Works directly with memory-mapped byte slices
Parallel Chunk Processing: Splits files into 4MB chunks processed via rayon
Lazy Field Extraction: Only parses fields needed for the query
Predicate Pushdown: Filters during scan, not after loading

Requirements

Rust 1.75+ (for building)
Claude Code or any MCP-compatible client

Contributing

See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Acknowledgments

Polars - Fast DataFrame library
rmcp - Rust MCP SDK
memchr - SIMD byte searching

forensic-log-mcp