vishalBrowserstack/mcp_server

MCP Log Analysis Server (Skeleton)

Overview

This is a skeleton implementation of a Model Context Protocol–style server that:

• Accepts raw or streamed log data
• Detects error patterns
• Produces suggestions combining pattern hints, Confluence documentation lookups, and Copilot (stub) AI reasoning
• Exposes a JSON-RPC-ish /mcp HTTP endpoint and a /stream WebSocket endpoint

Endpoints

POST /mcp

Request body example:

{
  "method": "log.analyze",
  "params": {
    "content": "2025-11-07T05:00:01Z ERROR Timed out waiting for resource\nTraceback ...",
    "mime": "text/plain"
  }
}

Response:

{
  "id": "...",
  "result": {
    "analysis": { "totalLines": 12, "errors": [ ... ] },
    "suggestions": [ ... ]
  }
}

Methods:

• log.analyze Analyze full log content
• log.patterns List registered patterns
• provider.searchConfluence Run a Confluence search
• provider.inspectCopilot Stub AI completion

WebSocket /stream

Messages:

• Client sends: {"type":"chunk","data":"line of log..."} repeatedly
• Client sends: {"type":"flush"} to trigger final analysis
• Server emits incremental detections and final analysis.

Configuration

Copy .env.example to .env and fill in:

CONF_BASE_URL=https://your-site.atlassian.net/wiki
CONF_USER=email@example.com
CONF_TOKEN=api_token
COPILOT_API_BASE=https://internal-copilot-endpoint.example.com
COPILOT_API_KEY=secret

Pattern Extension

Add new patterns in errorPatterns.ts, or expose a dynamic loader (e.g. YAML-driven) and call registry.add().

Copilot Integration

copilotProvider.ts is a stub. Replace complete() with:

• Internal enterprise endpoint
• Or alternative LLM (OpenAI, Anthropic, local model) while maintaining the provider interface.

Security & Privacy

• Basic redaction implemented in redact.ts; extend with more robust PII detection (regex + ML).
• Rate limiting token bucket available for provider calls.

Development

npm install
npm run dev

Next Steps

• Implement auth & tenancy
• Add richer structured parsing (JSON logs, ECS, OpenTelemetry)
• Add root-cause summarization aggregator
• Provide streaming SSE alternative
• Expand MCP handshake (capabilities advertisement)

Disclaimer

This code is a starting point and requires adaptation for production (error handling, resilience, secret management, tests).