Troubleshooting

Troubleshooting

Is the web app up? Open http://localhost:8725. The backend runs on :8724 and the MCP server on :8723.

Is the MCP server up? Open http://localhost:8723/health → {"status":"ok","version":"…"}.

A browser shows 405 at /mcp. Expected — /mcp is a POST-only MCP endpoint, not a web page. Use /health to check.

Claude Desktop doesn't show the tools. Fully quit and reopen Claude Desktop after editing the config; check the JSON is valid (no trailing commas); ensure the server/container is running. For the mcp-remote bridge, make sure Node.js is installed. See Connecting to Claude Desktop.

Port 8723 is already in use. Map a different host port (-p 9000:8723) or set MCP_HTTP_PORT.

Version reads unknown. The loaded dataset has no version metadata. The bundled channels include it, so this only appears with a custom SF_GAME_DATA_PATH dataset that lacks gameVersion.

Foreman chat

The foreman replies with 429 … exceeded your current quota. This comes from your LLM provider, not FICSIT Foreman — your request reached them and they declined it for billing reasons. API usage is prepaid and separate from any ChatGPT / Claude.ai subscription. Add credit and check your spend limit isn't $0:

• OpenAI: platform.openai.com → Settings → Billing.
• Anthropic: console.anthropic.com → Billing.

It can take a few minutes to propagate after adding credit. See Getting an OpenAI API key / Getting an Anthropic API key.

The foreman replies with 401 / "invalid API key" / an authentication error. The key is wrong, revoked, or doesn't match the selected provider (e.g. an OpenAI key with the provider set to Anthropic). In the web app's Settings, check the provider matches the key and that the key was pasted in full.

The foreman replies with "No LLM API key…". No key is configured. Either paste your own key in the web app's Settings, or set a server key (LLM_API_KEY, or ANTHROPIC_API_KEY for the default provider). See Configuration.

The first message after starting fails to use game data. If the backend started before the MCP server was listening, it logs Could not reach MCP server … will retry on first use. This is expected and self-heals — just send the message again and it reconnects.