Connect Codex to KabuForge

Install KabuForge in a Python 3.12+ environment and inspect kabuforge doctor. These examples use the main-branch connection contract. Replace the executable and workspace with your absolute paths. Market data is not bundled.

Run this where the Codex CLI is available. This example uses Windows paths. On macOS/Linux use /absolute/path/to/venv/bin/kabuforge and replace the workspace with its actual path.

codex mcp add kabuforge -- "C:/path/to/venv/Scripts/kabuforge.exe" --locale en_US mcp --workspace "C:/path/to/agent_workspace"

Instead of the registration command, merge this table into ~/.codex/config.toml or project-scoped .codex/config.toml. Do not register the same server twice.

[mcp_servers.kabuforge]
command = "C:/path/to/venv/Scripts/kabuforge.exe"
args = ["--locale", "en_US", "mcp", "--workspace", "C:/path/to/agent_workspace"]

codex mcp list

After checking registration, open a new Codex session. The terminal UI exposes active servers through /mcp. Once tool discovery succeeds, ask for capabilities and registered inputs first.

Use KabuForge read-only tools to inspect capabilities and registered input files, then tell me what is missing before starting research.

Place research files inside the specified workspace. Research tasks may write receipts there. Paper mutations are disabled by default; real order submission and cancellation are unsupported.

Check executable paths, the Python environment and workspace. stdout is reserved for MCP messages; inspect stderr and receipts for errors. This page provides configuration examples rather than claiming completed connection tests in every environment.

stdio connection details · AI assistant connections · Official OpenAI MCP guide