Why do tares factory crew up and tares factory spec stop and tell me Tares does not answer on a local setup?

Tares stops these commands on a local setup because the crew stations reach Tares only through the MCP endpoint, and that endpoint is not answering. Run tares up and tares mcp --transport streamable-http side by side, and both commands go through. With auth on, set TARES_AUTH_TOKEN before you start the MCP endpoint, and set TARES_FACTORY_MCP_URL if the endpoint runs somewhere else.

How it works

The software factory for Claude Code runs the crew as Claude Code sessions on your machine. Tares holds the plan, the tickets, the reviews and your answers. Every station talks to that record through one door, the MCP endpoint, so tares factory crew up and tares factory spec check that the door answers before they start anything.

On a local setup two processes make this work:

• tares up runs Tares itself, with the console on http://127.0.0.1:8787.
• tares mcp --transport streamable-http runs the MCP endpoint every crew session shares, at http://127.0.0.1:8788/mcp.

If you started only tares up, Tares is healthy but the stations have no way in, so the commands stop and tell you Tares does not answer. The same happens if the MCP endpoint has stopped, or if it is running at a different address from the one the sessions try.

The commands find your Tares in this order: TARESD_URL and TARES_AUTH_TOKEN when set, then the workspace that tares login saved, then the local Tares at http://127.0.0.1:8787. The sessions use http://127.0.0.1:8788/mcp for a local Tares, or the workspace address plus /mcp for a remote one.

Set it up

1. Start Tares in the first terminal.

   tares up

2. Open a second terminal and start the MCP endpoint. When auth is on, set TARES_AUTH_TOKEN to your token in that terminal first, then start the endpoint. Setting it afterwards does not help a process that is already running.

   # with auth on, set TARES_AUTH_TOKEN to your token before this line
   tares mcp --transport streamable-http     # http://127.0.0.1:8788/mcp

3. Leave both running. Use a third terminal for the crew.

4. If the MCP endpoint runs on another host or port, set TARES_FACTORY_MCP_URL to its address in the terminal where you run the crew commands. The default for a local Tares is http://127.0.0.1:8788/mcp.

5. Bring up the crew, then plan your project.

   tares factory crew up
   tares factory spec "what you want to build"

   tares factory spec comes after tares factory crew up.

6. If a command still stops, run tares factory doctor. It checks everything the crew needs and tells you which Tares it used, so you can see whether it reached the local one, the one from TARESD_URL, or your saved workspace.

If you use Tares Cloud instead, sign in with tares login and pick your workspace in the browser. There is no MCP endpoint to start yourself in that case. The full order of steps is on the Get started page.

What you get

• tares factory crew up starts the orchestrator, the reviewer and the builders, and the crew works on every project you hand it.
• tares factory spec opens Claude Code with the planning interview and writes the plan into a Tares project.
• Sessions share one MCP endpoint, so every station sees the same plan, tickets, reviews and answers.
• The record stays on Tares. If a station or the crew restarts, it carries on where it stopped.
• tares factory doctor gives you a quick check of the setup whenever something looks off.

When something else fits

• If you do not want to run two local processes, use Tares Cloud. You sign in with tares login and the crew uses your workspace, with no local MCP endpoint to start.
• If the message persists with both processes running, check the address first. Set TARES_FACTORY_MCP_URL to the right one, and make sure TARES_AUTH_TOKEN matches the token the endpoint was started with.

Related