Logfire MCP Server

What is Logfire MCP Server about?

The server exposes Pydantic Logfire telemetry—traces and metrics—through a set of RPC‑style tools. LLMs can call these tools to fetch recent exceptions, execute custom SQL against the Logfire DataFusion database, generate direct links to trace visualisations, and inspect the underlying schema.

How to use Logfire MCP Server?

Install uv (the recommended Python package manager).
Obtain a Logfire read token from the Logfire project settings.
Run the server with uvx logfire-mcp@latest and supply the token via the LOGFIRE_READ_TOKEN environment variable or the --read-token flag.
Configure your MCP client (Cursor, Claude Desktop, Cline, VS Code, Zed, etc.) using the examples provided—usually by setting the command to uvx with logfire-mcp@latest and the token in env or as an argument.

Key Features

find_exceptions_in_file – Retrieve up to 10 recent exceptions for a specific file within a configurable time window.
arbitrary_query – Execute any SQL query against the Logfire DataFusion store, constrained by a look‑back period.
logfire_link – Generate a URL that opens the trace directly in the Logfire UI.
schema_reference – Access the full OpenTelemetry schema for traces and metrics.
Environment‑based configuration – Support for LOGFIRE_READ_TOKEN and optional LOGFIRE_BASE_URL for self‑hosted deployments.

Use Cases

Debugging – Quickly surface recent errors in a particular module and jump to the full trace view.
Operational analytics – Run custom aggregations (e.g., error counts per service) without leaving the LLM chat.
Incident response – Generate shareable trace links for stakeholders during on‑call.
Exploratory querying – Data scientists can prototype SQL queries against live telemetry.

FAQ

Q: Do I need to run the server manually? A: Only if your MCP client does not manage servers for you. Clients like Cursor or Claude Desktop can launch the server automatically using the provided configuration.

Q: What is the maximum look‑back period? A: Both tools accept an age argument up to 7 days (10,080 minutes).

Q: How do I point the server at a self‑hosted Logfire instance? A: Set LOGFIRE_BASE_URL in the environment or use the --base-url flag when starting the server.

Q: Which language is the server written in? A: It is a Python package distributed via uvx.

Q: Is the server open‑source? A: Yes, under the MIT License.

Pydantic Logfire MCP Server

This repository contains a Model Context Protocol (MCP) server with tools that can access the OpenTelemetry traces and metrics you've sent to Pydantic Logfire.

This MCP server enables LLMs to retrieve your application's telemetry data, analyze distributed traces, and make use of the results of arbitrary SQL queries executed using the Pydantic Logfire APIs.

Available Tools

find_exceptions_in_file - Get the details about the 10 most recent exceptions on the file.
- Arguments:
  - filepath (string) - The path to the file to find exceptions in.
  - age (integer) - Number of minutes to look back, e.g. 30 for last 30 minutes. Maximum allowed value is 7 days.
arbitrary_query - Run an arbitrary query on the Pydantic Logfire database.
- Arguments:
  - query (string) - The query to run, as a SQL string.
  - age (integer) - Number of minutes to look back, e.g. 30 for last 30 minutes. Maximum allowed value is 7 days.
logfire_link - Creates a link to help the user to view the trace in the Logfire UI.
- Arguments:
  - trace_id (string) - The trace ID to link to.
schema_reference - The database schema for the Logfire DataFusion database.

Setup

Install `uv`

The first thing to do is make sure uv is installed, as uv is used to run the MCP server.

For installation instructions, see the uv installation docs.

If you already have an older version of uv installed, you might need to update it with uv self update.

Obtain a Pydantic Logfire read token

In order to make requests to the Pydantic Logfire APIs, the Pydantic Logfire MCP server requires a "read token".

You can create one under the "Read Tokens" section of your project settings in Pydantic Logfire: https://logfire.pydantic.dev/-/redirect/latest-project/settings/read-tokens

[!IMPORTANT] Pydantic Logfire read tokens are project-specific, so you need to create one for the specific project you want to expose to the Pydantic Logfire MCP server.

Manually run the server

Once you have uv installed and have a Pydantic Logfire read token, you can manually run the MCP server using uvx (which is provided by uv).

You can specify your read token using the LOGFIRE_READ_TOKEN environment variable:

LOGFIRE_READ_TOKEN=YOUR_READ_TOKEN uvx logfire-mcp@latest

You can also set LOGFIRE_READ_TOKEN in a .env file:

LOGFIRE_READ_TOKEN=pylf_v1_us_...

NOTE: for this to work, the MCP server needs to run with the directory containing the .env file in its working directory.

or using the --read-token flag:

uvx logfire-mcp@latest --read-token=YOUR_READ_TOKEN

[!NOTE] If you are using Cursor, Claude Desktop, Cline, or other MCP clients that manage your MCP servers for you, you do NOT need to manually run the server yourself. The next section will show you how to configure these clients to make use of the Pydantic Logfire MCP server.

Base URL

If you are running Logfire in a self hosted environment, you need to specify the base URL. This can be done using the LOGFIRE_BASE_URL environment variable:

LOGFIRE_BASE_URL=https://logfire.my-company.com uvx logfire-mcp@latest --read-token=YOUR_READ_TOKEN

You can also use the --base-url argument:

uvx logfire-mcp@latest --base-url=https://logfire.my-company.com --read-token=YOUR_READ_TOKEN

Configuration with well-known MCP clients

Configure for Cursor

Create a .cursor/mcp.json file in your project root:

{
  "mcpServers": {
    "logfire": {
      "command": "uvx",
      "args": ["logfire-mcp@latest", "--read-token=YOUR-TOKEN"]
    }
  }
}

The Cursor doesn't accept the env field, so you need to use the --read-token flag instead.

Configure for Claude code

Run the following command:

claude mcp add logfire -e LOGFIRE_READ_TOKEN=YOUR_TOKEN -- uvx logfire-mcp@latest

Configure for Claude Desktop

Add to your Claude settings:

{
  "command": ["uvx"],
  "args": ["logfire-mcp@latest"],
  "type": "stdio",
  "env": {
    "LOGFIRE_READ_TOKEN": "YOUR_TOKEN"
  }
}

Configure for Cline

Add to your Cline settings in cline_mcp_settings.json:

{
  "mcpServers": {
    "logfire": {
      "command": "uvx",
      "args": ["logfire-mcp@latest"],
      "env": {
        "LOGFIRE_READ_TOKEN": "YOUR_TOKEN"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Configure for VS Code

Make sure you enabled MCP support in VS Code.

Create a .vscode/mcp.json file in your project's root directory:

{
  "servers": {
    "logfire": {
      "type": "stdio",
      "command": "uvx", // or the absolute /path/to/uvx
      "args": ["logfire-mcp@latest"],
      "env": {
        "LOGFIRE_READ_TOKEN": "YOUR_TOKEN"
      }
    }
  }
}

Configure for Zed

Create a .zed/settings.json file in your project's root directory:

{
  "context_servers": {
    "logfire": {
      "source": "custom",
      "command": "uvx",
      "args": ["logfire-mcp@latest"],
      "env": {
        "LOGFIRE_READ_TOKEN": "YOUR_TOKEN"
      },
      "enabled": true
    }
  }
}

Example Interactions

Get details about exceptions from traces in a specific file:

{
  "name": "find_exceptions_in_file",
  "arguments": {
    "filepath": "app/api.py",
    "age": 1440
  }
}

Response:

[
  {
    "created_at": "2024-03-20T10:30:00Z",
    "message": "Failed to process request",
    "exception_type": "ValueError",
    "exception_message": "Invalid input format",
    "function_name": "process_request",
    "line_number": "42",
    "attributes": {
      "service.name": "api-service",
      "code.filepath": "app/api.py"
    },
    "trace_id": "1234567890abcdef"
  }
]

Run a custom query on traces:

{
  "name": "arbitrary_query",
  "arguments": {
    "query": "SELECT trace_id, message, created_at, attributes->>'service.name' as service FROM records WHERE severity_text = 'ERROR' ORDER BY created_at DESC LIMIT 10",
    "age": 1440
  }
}

Examples of Questions for Claude

"What exceptions occurred in traces from the last hour across all services?"
"Show me the recent errors in the file 'app/api.py' with their trace context"
"How many errors were there in the last 24 hours per service?"
"What are the most common exception types in my traces, grouped by service name?"
"Get me the OpenTelemetry schema for traces and metrics"
"Find all errors from yesterday and show their trace contexts"

Getting Started

First, obtain a Pydantic Logfire read token from: https://logfire.pydantic.dev/-/redirect/latest-project/settings/read-tokens

Run the MCP server:

uvx logfire-mcp@latest --read-token=YOUR_TOKEN

Configure your preferred client (Cursor, Claude Desktop, or Cline) using the configuration examples above
Start using the MCP server to analyze your OpenTelemetry traces and metrics!

Contributing

We welcome contributions to help improve the Pydantic Logfire MCP server. Whether you want to add new trace analysis tools, enhance metrics querying functionality, or improve documentation, your input is valuable.

For examples of other MCP servers and implementation patterns, see the Model Context Protocol servers repository.

License

Pydantic Logfire MCP is licensed under the MIT License. This means you are free to use, modify, and distribute the software, subject to the terms and conditions of the MIT License.