Oluwatunmise-olat/mcp-server-logs-sieve

mcp-server-logs-sieve is an MCP server that lets AI assistants query logs directly from your observability backend.
Ask debugging questions in plain language, and let Logs Sieve pull the exact logs and context for you.
Why this exists
During incidents, a lot of time goes into repetitive steps:
- finding the right log source
- narrowing the time window
- spotting recurring failure patterns
- following one trace across services
Logs Sieve packages those workflows into four tools: query_logs, summarize_logs, trace_request, and list_log_sources.
Supported providers
- Google Cloud Logging (gcp)
- AWS CloudWatch Logs (aws)
- Azure Monitor Logs (azure)
- Grafana Loki (loki)
- Elasticsearch (elasticsearch)
Prerequisites
> **⚠️ Node.js 20 or above is required.**
>
> This package uses ES modules and dependencies that need Node.js 20+. If you see ERR_MODULE_NOT_FOUND errors, check your Node version with node -v and upgrade if needed.
>
> This applies whether you run via npx, install globally, or use from source.
Quick MCP config
Example .mcp.json:
If you are running from this repo source directly, you can still use node ./bin/mcp-server-logs-sieve.js --provider gcp.
**First-time setup:** On first use, npx needs to download and install the package (~40s). Claude Code may time out waiting for this. Run the following once in your terminal to prime the cache before adding it to your MCP config:
After that, Claude Code will start the server instantly from the npx cache.
After updating config, restart your MCP client and confirm with /mcp.
Provider auth and env vars
See provider-specific setup and auth docs:
- GCP
- AWS
- Azure
- Loki
- Elasticsearch
For Elasticsearch API compatibility headers:
CLI usage
The package also includes a CLI for direct terminal usage:
Documentation
- Docs home
- Getting started
- MCP client setup (Claude Code, Claude Desktop, Cursor, Windsurf)
- Tool reference