Log Checker

You run a self-contained MCP server that analyzes logs locally, using semantic search, error pattern detection, and fast vector search with FAISS. It operates entirely offline, preserving privacy while delivering rapid insights from your log data.

How to use

Connect to the Log Checker MCP Server with a client that supports MCP. You have two practical ways to run and query it locally: the UVX-based cloud-free flow for quick setup, or running the Python server directly for full control. In either case, you can fetch, vectorize, and query your logs to discover meaning, detect error patterns, and summarize activity across time ranges.

Using the UVX flow is the recommended quick-start path. It installs the UVX launcher and configures a local MCP server entry that points to the MCP package in your logs project. This method keeps everything local and simple to start from Claude Desktop or other clients that support MCP. Alternatively, you can run the server directly with Python to have full control over the environment and direct interactions with the server script.

Example usage patterns you can perform after you have the MCP server running: fetch and chunk local log files, embed chunks to vectors, then query the vector store semantically and lexically. Ask for summaries of errors, detect recurring error types, and inspect timeframes and method calls associated with issues. The system provides automated clustering of errors and AI-assisted summaries to help you interpret complex log streams.

How to install

Prerequisites you need before installation: a modern Python runtime, pip for Python package installation, and a MCP client or launcher you plan to use (such as UVX) for stdio-based MCP connections.

# UVX quick-start (recommended)
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

# UVX MCP JSON example to register the server
# Save this as a config snippet for your Claude Desktop or MCP client
{
  "mcpServers": {
    "log-analyzer": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/suriya-ML/log-checker-mcp.git",
        "log-analyzer-mcp"
      ]
    }
  }
}

Environment and configuration

The server can be configured to tune embedding and analysis behavior, and it uses environment variables for credentials and model selection. Typical variables include AWS access keys for embedding models, the region, and model IDs for Bedrock if you enable cloud-backed embeddings or analysis. You also specify the folder where logs are stored and how large each chunk should be during processing.

AWS_ACCESS_KEY_ID=YOUR_AWS_ACCESS_KEY
AWS_SECRET_ACCESS_KEY=YOUR_AWS_SECRET_KEY
AWS_REGION=us-east-2
BEDROCK_EMBED_MODEL_ID=amazon.titan-embed-text-v2:0
BEDROCK_NOVA_MODEL_ID=amazon.nova-premier-v1:0
LOG_FOLDER=./logs
DEFAULT_CHUNK_SIZE=4096
DEFAULT_OVERLAP=1024

Troubleshooting

Common issues include missing vector data, authentication problems with Bedrock, or missing log chunks. Ensure you have vector data created by running the vectorization step before querying. Verify AWS credentials and Bedrock access if you rely on cloud embeddings. Check that logs exist in the configured folder and that file extensions match what the server expects (for example, .log or .txt).

If you need to debug, run the server in a way that surfaces logs to the console or a file for troubleshooting. For example, redirect standard error to a log file when starting the server to capture diagnostic messages.

Notes

This MCP server operates locally, prioritizing privacy and offline processing while providing powerful semantic search, error clustering, and summarization capabilities for logs.

Available tools

fetch_local_logs

Fetch and chunk log files from a local directory with configurable chunk size and overlap.

store_chunks_as_vectors

Vectorize log chunks into embeddings, with optional caching and parallel processing.

query_SFlogs

Query vectorized logs using semantic and lexical search, with error clustering and AI-driven summaries.