Command Palette

Search for a command to run...

UnylyUnyly
Browse all

Sakana

FreeNot checked

MCP server that wraps Sakana AI Scientist v2 so MCP Client (claude, cursor...) can act as a research director for autonomous scientific loops.

GitHubEmbed

About

MCP server that wraps Sakana AI Scientist v2 so MCP Client (claude, cursor...) can act as a research director for autonomous scientific loops.

README

FastMCP Version Ruff Built with Just

sakana-mcp

MCP server that wraps Sakana AI Scientist v2 so Cursor can act as a research director for autonomous scientific loops.

Documentation Map

  • README.md (this file): entrypoint and quick start
  • docs/README.md: full docs index
  • docs/architecture/README.md: system architecture and component boundaries
  • docs/researcher/README.md: researcher-facing workflow (ideate -> execute -> status -> review)
  • docs/operations/README.md: Docker, safety model, runbooks, and troubleshooting
  • research_library/README.md: sample research task catalog
  • prompts/README.md: reusable prompt templates
  • skills/FASTMCP_3_1_SKILL.md: FastMCP 3.1.0 operator skill card
  • docs/ONBOARDING.md: first-time setup (keys, Docker, vendor checkout) — read this before just serve

New here? Start with docs/ONBOARDING.md (15 min: keys → Docker → vendor → Go Live Checklist).

Core Guarantees

  • Tool-based orchestration: MCP tools map to ideation, execution, review, and status loops
  • Sandbox-first execution: experiment code runs in Docker to reduce host contamination risk
  • Deep-memory export: run artifacts are persisted in research_vault for downstream RAG indexing

Tool Surface

  • research_ideate: generate 5 hypotheses using AI-Scientist v2 ideation if available
  • research_execute: run experiment manager in Docker with num_workers and max_debug_depth
  • research_status: summarize buggy/non-buggy tree nodes from run artifacts
  • research_review: run VLM figure/caption/reference review on a produced PDF
  • research_library: browse built-in + custom sample research missions
  • research_workflow_plan: generate guided agentic step plan for a selected task

Stack

  • Backend: Python 3.12+, FastMCP 3.4.5 (stdio), FastAPI + uvicorn (dashboard :10863), pydantic v2, httpx, ruff + mypy gates, pytest smoke suite
  • Frontend: React 18 + Vite 5 + TailwindCSS 3 + TanStack Query + React Router (dashboard :10862), TypeScript strict, Biome 2.4 gate, Playwright e2e
  • Runtime: Docker Compose (sakana-mcp + GPU scientist-executor), Tauri shell (native/), MCPB bundle via fleet shim (just mcpb-pack)

Research Warehousing

All major events are appended to warehouse manifests under:

  • research_vault/warehouse/manifests/*.jsonl

Warehouse directories are auto-created under research_vault/warehouse:

  • runs, reviews, manuscripts, datasets, manifests, exports

Environment Variables

  • GEMINI_API_KEY: required for Gemini-backed ideation/review paths
  • S2_API_KEY: Semantic Scholar key for literature-heavy workflows
  • RESEARCH_VAULT_PATH: optional; defaults to ./research_vault
  • AI_SCIENTIST_V2_PATH: optional; defaults to ./vendor/ai-scientist-v2

Quick Start

git clone https://github.com/sandraschi/sakana-mcp
cd sakana-mcp
just

This opens an interactive dashboard showing all available commands. Run just bootstrap to install dependencies, then just serve or just dev to start.

Manual Setup

If you don't have just installed:

  1. Clone this repo and ensure vendor/ai-scientist-v2 exists.
  2. Set required environment variables.
  3. Build Docker images:
  • docker compose build
  1. Run MCP server locally:
  • python -m sakana_mcp.server

Safety Position

Docker is strongly recommended and treated as the default execution path for experiments because AI-Scientist v2 executes model-generated code.

🛡️ Industrial Quality Stack

This project adheres to SOTA 14.1 industrial standards for high-fidelity agentic orchestration:

  • Python (Core): Ruff for linting and formatting. Zero-tolerance for print statements in core handlers (T201).
  • Webapp (UI): Biome for sub-millisecond linting. Strict noConsoleLog enforcement.
  • Protocol Compliance: Hardened stdout/stderr isolation to ensure crash-resistant JSON-RPC communication.
  • Automation: Justfile recipes for all fleet operations (just lint, just fix, just dev).
  • Security: Automated audits via bandit and safety.

from github.com/sandraschi/sakana-mcp

Installing Sakana

This server has no published package — it is built from source. Open the repository and follow its README.

▸ github.com/sandraschi/sakana-mcp

FAQ

Is Sakana MCP free?

Yes, Sakana MCP is free — one-click install via Unyly at no cost.

Does Sakana need an API key?

No, Sakana runs without API keys or environment variables.

Is Sakana hosted or self-hosted?

Self-hosted: the server runs locally on your machine via the install command above.

How do I install Sakana in Claude Desktop, Claude Code or Cursor?

Open Sakana on unyly.org, pick your client tab (Claude Desktop, Claude Code, Cursor) and press Install — the config is generated automatically, no JSON editing.

Related MCPs

Compare Sakana with

Not sure what to pick?

Find your stack in 60 seconds

Author?

Embed badge for your README

Browse similar

All ai MCPs