tracehub-mcp: Giving AI Assistants a Real Query Interface Into Your LLM Traces

Description and use

tracehub-mcp enables AI assistants to query OpenTelemetry trace backends directly, removing the need to copy and paste JSON. It supports five backends, enforces HTTPS, safe query input, and bounded pagination, ensuring reliable and secure trace access.

Problem

Debugging LLM applications requires manually copying trace JSON from a backend and pasting it into an AI assistant, creating a tedious loop and limiting the assistant’s ability to query traces directly.

Outcome

tracehub-mcp eliminates the manual copy/paste loop, enabling AI assistants to query trace backends directly. It supports five backends with full implementations, enforces HTTPS, query‑injection safety, bounded pagination, and strict ID re‑verification, ensuring reliable and secure trace retrieval.

Workflow

• 1. Deploy tracehub-mcp server and configure it with a trace backend URL and credentials.
• 2. Configure the AI assistant (e.g., Claude) to use tracehub-mcp as an MCP client.
• 3. In the assistant chat, invoke MCP tools such as search_traces or get_llm_expensive_traces with desired query parameters.
• 4. The assistant receives structured trace data and can reason over it to answer questions like "compare the last five calls to this model".
• 5. Results are displayed directly in the chat, eliminating manual copy/paste and spreadsheet work.

Evidence

tracehub-mcp exists to delete that loop. It's an MCP (Model Context Protocol) server that gives Claude, Cursor, Windsurf, Gemini CLI, or any MCP client a direct query interface into your OpenTelemetry trace backend.

Source

Every backend implements the same BaseBackend interface, so every MCP tool works identically no matter which backend is configured.

Source

HTTPS-only enforcement. Both refuse to construct at all if the configured URL isn't https:// — a ValueError raised in __init__.

Source