Logfire

Exposes distributed traces and spans stored in Pydantic Logfire through an MCP tool interface that accepts natural language queries. The AsyncLogfireQueryClient handles HTTP communication with Logfire's REST API, translating user intent into structured queries against OpenTelemetry-formatted telemetry data. Integrates with FastMCP's tool registration system to expose query capabilities to LLM clients via JSON-RPC over stdio transport.

The find_exceptions_in_file tool queries Logfire for exceptions that occurred in a specific source file, returning stack traces with line numbers and context. Implements file-scoped exception filtering by querying OpenTelemetry exception spans and matching them against the provided file path. Results include full exception details, timestamps, and surrounding code context for rapid debugging.

The arbitrary_query tool exposes direct SQL access to Logfire's DataFusion-backed database, allowing users to execute custom queries against OpenTelemetry metrics, traces, and spans. Queries are executed via AsyncLogfireQueryClient's HTTP interface, with results returned as structured JSON. Enables power users and data analysts to perform complex aggregations, joins, and filtering beyond the scope of predefined tools.

The logfire_link tool generates direct URLs to specific traces in the Logfire web UI, enabling seamless navigation from LLM-assisted debugging back to the interactive Logfire dashboard. Takes a trace ID and constructs a properly formatted URL that opens the trace in Logfire's UI with full span visualization, metrics, and context. Implements URL construction logic that handles Logfire's project-scoped URL structure.

The schema_reference tool queries Logfire's DataFusion database to retrieve table definitions, column names, types, and metadata, enabling users to understand the structure of available telemetry data. Executes schema queries via AsyncLogfireQueryClient and returns structured metadata that helps users construct valid SQL queries. Supports both full schema dumps and targeted table/column lookups.

Implements the MCP server runtime using FastMCP framework, handling stdio transport, JSON-RPC message routing, and tool registration. The app_factory function creates a FastMCP instance with a lifespan context that initializes AsyncLogfireQueryClient on startup and manages its lifecycle. Implements proper async context management to ensure Logfire client is available for all tool invocations and cleaned up on shutdown.

Implements multi-source token resolution that checks command-line arguments, environment variables, and .env files to obtain Logfire read tokens. The main() CLI entry point uses this resolution logic to initialize AsyncLogfireQueryClient with proper credentials. Supports both explicit token passing and environment-based discovery, enabling flexible deployment across local development and production environments.