Logfire SDK CLI: SDK Command Line Interface Guide
Logfire comes with a CLI used for authentication and project management:
usage: logfire [-h] [--version] [--base-url BASE_URL | --region {us,eu}] ...
The CLI for Pydantic Logfire.
options:
-h, --help show this help message and exit
--version show the version and exit
global options:
--base-url BASE_URL the base URL for self-hosted Logfire instances (e.g.,
http://localhost:8080)
--region {us,eu} the region to use
commands:
auth Authenticate with Logfire.
clean Remove the contents of the Logfire data directory.
inspect Inspect installed packages and recommend packages that
might be useful.
whoami Show user authenticated username and the URL to your
Logfire project.
projects Project management for Logfire.
read-tokens Manage read tokens for a project
prompt Creates a prompt to be used with your favorite LLM. The
prompt assumes you are using Logfire MCP.
info Show versions of logfire, OS and related packages.
run Run Python scripts/modules with Logfire instrumentation
See https://logfire.pydantic.dev/docs/reference/cli/ for more detailed
documentation.
You need to be authenticated to use the Logfire.
To authenticate yourself, run the auth command in the terminal:
logfire auth
You will be prompted to select a data region (EU or US). To specify this
via the cli instead of interactively, use logfire --region eu auth or logfire --region us auth
Then you will be given the option to open logfire in your browser:
After pressing "Enter", you will be redirected to the browser to log in to your account.
Then, if you go back to the terminal, you’ll see that you are authenticated! 🎉
To log out and remove the locally stored credentials, run:
logfire auth logout
To clean most the files created by Logfire, run the following command:
logfire clean
The clean command doesn’t remove the logs, and the authentication information stored in the ~/.logfire directory.
To also remove the logs, you can run the following command:
logfire clean --logs
The gateway command runs a local OAuth proxy for the Logfire AI Gateway and can launch supported AI coding tools with short-lived credentials.
Install the optional dependencies before using it:
uv add "logfire[gateway]"
pip install "logfire[gateway]"
poetry add "logfire[gateway]"
Launch a supported integration through the proxy:
logfire gateway launch claude
You can also run only the proxy and configure a tool manually:
logfire gateway serve
Use --device-flow if browser callback authorization is not available, or pass --region eu / --region us before gateway to select the Logfire region.
The inspect command is used to identify the missing OpenTelemetry instrumentation packages in your project.
To inspect your project, run the following command:
logfire inspect
This will output the projects you need to install to have optimal OpenTelemetry instrumentation.
To print the versions of logfire, your operating system, and related (OpenTelemetry) packages (useful when reporting a bug) run:
logfire info
To show the currently authenticated user and the URL of the current Logfire project, run:
logfire whoami
If you have one or more tokens configured (e.g. via the LOGFIRE_TOKEN environment variable), this shows the project each token belongs to instead.
To check the projects you have access to, run the following command:
logfire projects list
You’ll see something like this:
❯ logfire projects list
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┓
┃ Organization ┃ Project ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━┩
│ Kludex │ backend │
│ Kludex │ worker │
└──────────────┴────────────────┘
To use an already created project, run the following command:
logfire projects use <project-name>
For example, to use the backend project, you can run:
logfire projects use backend
To create a new project, run the following command:
logfire projects new <project-name>
Follow the instructions, and you’ll have a new project created in no time! 🥳
To create a read token for a project and print it to stdout, run:
logfire read-tokens --project <org>/<project> create
Because the token is printed to stdout, this composes well with other tools, e.g. configuring the Logfire MCP server:
claude mcp add logfire -e LOGFIRE_READ_TOKEN=$(logfire read-tokens --project <org>/<project> create) -- uvx logfire-mcp@latest
To run a Python script or module with Logfire instrumentation enabled automatically for all installed packages that have an available OpenTelemetry instrumentation, run:
logfire run script.py
# or run a module, forwarding any arguments after it:
logfire run -m my_module --my-arg
By default a summary box is printed to stderr showing which packages were instrumented; disable it with --no-summary. Use --exclude to skip instrumenting specific packages:
logfire run --exclude sqlalchemy,fastapi script.py
To generate a prompt for an LLM to investigate an issue in your project (assuming the Logfire MCP server is configured), run:
logfire prompt "why are my requests slow?"
Use --claude, --codex, or --opencode to verify (and set up) the MCP configuration for the respective coding tool, --update to replace an existing Logfire MCP server configuration, and --project <org>/<project> to select the project.