⬡ Sato Hub MCP

Query the scored, daily-updated index of the onchain agent stack — live, from your agent.

npx -y @satohub/mcp · Streamable HTTP · 35 tools · four can change Sato Hub state · no API key

satohub.ai/mcp · the index · framework plugins · satohub.ai

• What this is. @satohub/mcp is the stdio shim for the hosted Sato Hub MCP server at https://satohub.ai/api/mcp. The server exposes 35 tools over a daily-rebuilt, evidence-scored index of onchain-agent tooling, plus Preflight, build planning and non-custodial routing. Use the remote Streamable HTTP endpoint directly when your client supports it; use this package when it requires stdio. No API key is required.
• What it can change. Thirty-one tools are read-only. Four can change Sato Hub state: onchain_agent_watch, onchain_agent_submit_project, onchain_agent_register_agent, and onchain_agent_swap in build-tx mode. Sato Hub never signs, holds or broadcasts funds.
• What a Sato Score is. A 0–100 measure of how open, active and verifiable a project is, computed from evidence only. It is not a safety, quality, security or returns grade, and self-reported is never treated as verified. Methodology →
• Non-custodial. Nothing here signs a transaction, holds a key or moves funds. Route and swap tools return unsigned objects for your own signer, and every refusal names its rule.

Endpoint

POST https://satohub.ai/api/mcp

Model Context Protocol over Streamable HTTP. If your client speaks HTTP, use this URL directly and skip the package.

Connect

Claude Code

claude mcp add --transport http satohub https://satohub.ai/api/mcp

Claude Desktop (claude_desktop_config.json) · Cursor (.cursor/mcp.json) · Windsurf (mcp_config.json)

{
  "mcpServers": {
    "satohub": { "command": "npx", "args": ["-y", "@satohub/mcp"] }
  }
}

VS Code, Codex, any Streamable HTTP client

{ "mcpServers": { "satohub": { "type": "http", "url": "https://satohub.ai/api/mcp" } } }

The shim wraps mcp-remote against the hosted endpoint. Extra arguments are forwarded to it; SATOHUB_MCP_ENDPOINT overrides the endpoint.

The 35 tools

All names are prefixed onchain_agent_.

| Layer | Tools | |---|---| | Find and judge | search_resources · get_resource · compare_listings · recommend_stack · get_deploy_spec · get_score_methodology · get_listing_history | | Check before you act | preflight · watch ▲ | | Route (quotes, unsigned) | route_swap · swap ▲ · route_agent · route_launch · route_lp | | Build | build_plan · scaffold_plan | | Agent economy & numbers | get_agent_economy · get_trend · explain_number · get_metrics | | Agents & skills | search_agents · get_agent_passport · search_skills · search_listings · inspect_listing | | Stay current | recent_changes · get_changes · get_news · get_data_freshness | | Reference | list_categories · list_chains · list_wiki_pages · get_wiki_page | | File | submit_project ▲ · register_agent ▲ — a submission or a self-reported registration, never a listing and never a verification |

▲ Thirty-one tools are read-only. Four can change Sato Hub state: watch creates a monitor, submit_project queues a submission, register_agent creates or pends a self-reported passport, and swap in build-tx mode records a receipt for an unsigned transaction. None signs, holds or broadcasts funds.

Every record carries sato_url (its canonical page) and, where scored, verify_url (the Sato Score report). Cite the sato_url when you surface a record so the reader can check its current state. Preflight verdicts and route decisions leave the server with a detached Ed25519 signature — the scheme.

Try it in 10 seconds

Ask your MCP-connected agent:

"Search the Sato Hub index for Solana trading MCP servers with a Sato Score above 70, run Preflight on the top one, and give me its deploy spec."

No MCP client? The raw feed works anywhere:

curl -s https://satohub.ai/api/export/index.json | jq '.resources[] | select(.trust_score > 80) | .name'

What the numbers mean

verification_status separates self-reported from verified. In agent-economy data null means unknown, never zero, and rows are never summed across venues, chains or stages. A Preflight unknown is a refusal to guess, not a pass.

Maintained by Sato Hub — the agent builder hub for crypto. Server: mcp-handler on Next.js. MIT © Sato Hub (this package and documentation); index data CC-BY-4.0, data by satohub.ai.