@plugsky/mcp — Plugsky over the Model Context Protocol

Use all 36 Plugsky models, your RAG knowledge, the platform tools and document ingestion from any MCP client: Claude Desktop, Cursor, VS Code, Windsurf, Zed, ChatGPT (connectors).

Two ways to connect:

| Transport | Endpoint | Auth | |---|---|---| | Remote (recommended) — Streamable HTTP | https://plugsky.com/mcp | Authorization: Bearer sk-live-… | | Local — stdio | npx -y @plugsky/mcp | env PLUGSKY_API_KEY (+ PLUGSKY_EMAIL/PLUGSKY_PASSWORD for dashboard tools) |

The remote endpoint speaks JSON-RPC 2.0 over POST (stateless Streamable HTTP, protocols 2024-11-05 → 2026-07-28) and exposes plugsky_chat, plugsky_models and plugsky_fusion. The local server additionally exposes the dashboard tools (RAG, usage, platform tools, document ingestion) through your account login.

Quick start (local)

PLUGSKY_API_KEY=sk-live-… npx -y @plugsky/mcp            # all tool groups
PLUGSKY_API_KEY=sk-live-… npx -y @plugsky/mcp --groups=rag

Dashboard tools (RAG, usage, tools, files) also need your account login:

PLUGSKY_API_KEY=sk-live-… [email protected] PLUGSKY_PASSWORD=… npx -y @plugsky/mcp

Tool groups

| Group | Tools | Auth | |---|---|---| | chat | plugsky_chat | API key | | models | plugsky_models | API key | | rag | plugsky_rag_collections, plugsky_rag_query | + dashboard login | | usage | plugsky_usage | + dashboard login | | tools | plugsky_tools_list, plugsky_tools_run | + dashboard login (Builder+) | | files | plugsky_documents_list, plugsky_documents_ingest | + dashboard login |

--groups=chat,rag (or PLUGSKY_GROUPS=chat,rag) filters the exposed tools.

Client configuration

Claude Desktop / Cursor / Windsurf (local)

{
  "mcpServers": {
    "plugsky": {
      "command": "npx",
      "args": ["-y", "@plugsky/mcp"],
      "env": {
        "PLUGSKY_API_KEY": "sk-live-…",
        "PLUGSKY_EMAIL": "[email protected]",
        "PLUGSKY_PASSWORD": "…"
      }
    }
  }
}

Claude / Cursor / ChatGPT (remote)

{
  "mcpServers": {
    "plugsky": {
      "url": "https://plugsky.com/mcp",
      "headers": { "Authorization": "Bearer sk-live-…" }
    }
  }
}

In ChatGPT: Settings → Connectors → Add MCP server (developer mode) → enter the URL https://plugsky.com/mcp and your API key as the bearer token.

VS Code

{
  "servers": {
    "plugsky": {
      "command": "npx",
      "args": ["-y", "@plugsky/mcp"],
      "env": { "PLUGSKY_API_KEY": "sk-live-…" }
    }
  }
}

Env vars

| Var | Default | Purpose | |---|---|---| | PLUGSKY_API_KEY | — (required) | sk-live-… key for /v1 inference + models | | PLUGSKY_EMAIL / PLUGSKY_PASSWORD | — | dashboard login for RAG / usage / tools / files | | PLUGSKY_GROUPS | chat,models,rag,usage,tools,files | tool group filter | | PLUGSKY_BASE_URL | https://api.plugsky.com/v1 | API base | | PLUGSKY_SITE_URL | https://plugsky.com | dashboard base (RAG, docs) |

Why Plugsky MCP?

• Free inference tier for agents — plugsky-micro / plugsky-lite cost nothing.
• 36 models, one key — reasoning, vision, coder, long-context, fusion.
• Your own knowledge — vector search over your RAG collections, with scores.
• End-to-end documents — ingest a PDF/image/scan by URL and query it right after.
• No per-token markup — flat plans, unlimited usage under fair-use RPM.

MIT © Plugsky