What do I need installed and signed in before a crew can work in a Python backend?

Before a crew works in a Python backend, Tares needs Claude Code signed in, gh signed in with write access, and git, jq and curl installed. Claude Code must be on a plan that allows background sessions, because every crew member is a background Claude Code session. A second-model CLI, the Codex CLI, is optional. Run tares factory doctor to check all of it.

How it works

The crew always runs on your machine as Claude Code sessions, so the prerequisites are local tools plus two sign-ins. The software factory for Claude Code uses each one for a different job:

• Claude Code, signed in, on a plan that allows background sessions. The orchestrator, reviewer and builders use your Claude Code login. There is no separate model API key to create.
• gh, signed in with write access to the repos the crew works in. Builders push branches and open pull requests with your gh login, so the account behind it needs write access to your Python repos.
• git, jq and curl. These must be installed on the machine that runs the crew.
• Optional: the Codex CLI. Install it if you want a second model to check every commit. The first tares factory crew up asks whether to turn this on.

Tares also needs its own access to the repos, separate from gh. It reads the pull request, CI and merge state of every ticket itself. On a local Tares you add a GitHub token or App under Settings > GitHub. On Tares Cloud you choose Connect GitHub in your workspace settings and pick the repos.

For a Python backend, the crew also looks for a Python language server. Builders and the reviewer search less when Claude Code can find a definition, its references and its callers. Tares detects Python from pyproject.toml, setup.py, setup.cfg, Pipfile or requirements.txt, at the repo root or one folder down. A server counts when its plugin is enabled for you at user scope and its binary is on your PATH. The install uses npm for the server, then a Claude Code plugin install. This part is optional, and a missing server does not stop the crew.

Set it up

1. Check the prerequisites. Make sure Claude Code is signed in on a plan that allows background sessions, and that git, jq and curl are installed. Sign in to gh with an account that has write access to your Python repos:

   gh auth login

2. Install the CLI:

   uv tool install tares     # or: brew install glassflow/tap/tares

3. Choose where Tares runs.

   • Local: start Tares with tares up, then in a second terminal start the MCP endpoint every crew session shares. Crew sessions reach Tares only through this endpoint, so crew up and spec stop and tell you when it does not answer.

     tares up
     tares mcp --transport streamable-http

   • Tares Cloud: run tares login, pick your workspace in the browser and choose Connect. There is no MCP endpoint to start.

     tares login

4. Give Tares access to GitHub, in Settings > GitHub for a local Tares or Connect GitHub on Tares Cloud. See Get started for both paths.

5. Bring up the crew:

   tares factory crew up

   The first run offers to open Claude Code in ~/factory so you can trust the folder once. It then asks how far the crew may go, including whether a second model checks every commit. If the Python language server is missing, it lists what is missing and offers to install it. It installs only when you say yes. --yes and a closed input install nothing.

6. Check everything the crew needs, then plan your first project from inside the Python repo:

   tares factory doctor
   tares factory spec "what you want to build"

What you get

• A crew of background Claude Code sessions (orchestrator, reviewer and builders) that works on your own login, with no extra model API key.
• Builders that push and open pull requests on your Python repos with your gh login, and Tares tracking each ticket’s pull request, CI and merge state.
• A doctor report with one line per requirement and, for each problem, the command that fixes it. It also shows the key the stations use and each station’s Claude Code version.
• Language server checks for your Python repos, so builders and the reviewer can search by symbol.
• An optional second-model check on every commit if you install the Codex CLI.

When something else fits

• If you do not want the crew to run in the background on your machine, work on a project in your own Claude Code session with tares factory session. The same prerequisites apply, except you drive the work yourself.
• If you have no Claude Code plan that allows background sessions, the crew cannot start, because every station is a Claude Code session.

Related