MCP Server

Plug deepseekbase data into any MCP-capable AI client — DeepSeek Harness, Claude, Cursor, Cline. Configure once and your assistant can look up DeepSeek prices, plugins, model specs and industry kits instead of relying on training memory.

What is this? MCP (Model Context Protocol) is the standard way AI clients connect to external data. We run an MCP server, so when your assistant is asked "how much does DeepSeek cost" it can query live data rather than guess.

Endpoint

https://deepseekbase.com/mcp

Protocol: JSON-RPC 2.0 over HTTP POST (Streamable HTTP). Read-only, no API key.

Five tools

Tool	What it answers
`price_query`	Official prices plus provider quotes, with quantization and the price spread
`plugin_search`	8,559 DSH plugins with downloads, risk level, permissions and red lines
`model_specs`	Layers, dimensions, context, and derived models for a domain
`industry_kit`	Five-layer kit per industry: models, toolchain, reference, hardware, pitfalls
`official_resources`	Official repositories, API endpoints and runnable commands

Connect (DSH)

{
  "mcpServers": {
    "deepseekbase": {
      "url": "https://deepseekbase.com/mcp"
    }
  }
}

Connect (Claude Desktop / Cursor)

{
  "mcpServers": {
    "deepseekbase": {
      "url": "https://deepseekbase.com/mcp"
    }
  }
}

Try it

curl -X POST https://deepseekbase.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Call a tool

curl -X POST https://deepseekbase.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0", "id": 2, "method": "tools/call",
    "params": {
      "name": "price_query",
      "arguments": {"model": "v4.1-flash", "limit": 5}
    }
  }'

Principles

Read-only — it cannot change anything, write files or execute code.
No key — unlike the official API, no authentication.
Timestamped — every dataset carries verified_at so the assistant knows how fresh it is.
Attributed — results include links to official sources.

Datasets · Plugins · Official resources