Are you the author? Sign in to claim
A testbed for auditable, agent-operated systems: a financial terminal an AI drives end-to-end over MCP, with hard paper/
A local financial terminal your AI agent operates for you. Market data, backtests, paper books, news — driven in plain language, running entirely on your machine.

Needs uv and Python 3.12. No clone, no build, no server to start:
claude mcp add otto -- uvx --from git+https://github.com/0Smallcat0/otto otto-mcp
Then ask for things:
Want the dashboard as well: npm --prefix frontend install && npm --prefix frontend run build,
then uv run otto → http://127.0.0.1:8765/
It keeps score. The agent records dated calls — a stance, the reasoning, the price level that would prove it wrong, a horizon — and every call is later graded against the price the market actually printed. Calls without reasoning are refused rather than guessed, moves inside a flat band don't count as skill, and a call graded long after it matured is excluded because it measured a window its thesis never claimed. The hit rate is whatever the market says it is.
"An AI can operate it" is a benchmark here, not a tagline. A real headless agent gets 20 plain-language tasks in hermetic sandboxes, graded programmatically on terminal state and artifacts — never by an LLM judge:
| Model | Passed | Avg turns |
|---|---|---|
claude-sonnet-5 | 20 / 20 | 6.5 |
claude-haiku-4-5 | 19 / 20 | 6.8 |
Safety tasks grade refusal: asking for a live order must leave state unchanged. Live trading, credential entry and code execution aren't switched off — they're unreachable through the surface the agent has.
One typed contract (135 actions across 16 routes) is the single source of truth; the MCP tools, the UI capability catalog and the eval suite are all derived from it. 636 tests on Windows + Linux CI.
python -m pytest -q && python -m ruff check .
MIT.
Run Claude Code as an MCP server so any agent can delegate coding tasks to it
Browser automation using accessibility snapshots instead of screenshots
Google's universal MCP server supporting PostgreSQL, MySQL, MongoDB, Redis, and 10+ databases
Official GitHub integration for repos, issues, PRs, and CI/CD workflows