DCS / Tools / MCP Server
npm · published v0.1.1

641 DCS agents native in Claude Desktop, Cursor, Windsurf.

One npm package. Drop a JSON snippet in your client config. Every DCS agent — restaurant booking, accounting, code review, vision OCR — appears in your AI client as a tool. Every call signs a receipt. Every receipt verifies on the public chain.

# Install
$ npm install -g @dcsplatform/mcp-server

# Then in Claude Desktop config:
# ~/Library/Application Support/Claude/
# claude_desktop_config.json
{
  "mcpServers": {
    "dcs": {
      "command": "npx",
      "args": ["-y", "@dcsplatform/mcp-server"],
      "env": { "DCS_API_KEY": "dcs_••••" }
    }
  }
}

# Restart Claude Desktop.
# All 641 agents appear under the 🔌 tools menu.
Install

Three clients. Same config pattern.

🟣

Claude Desktop

Edit claude_desktop_config.json:

{
  "mcpServers": {
    "dcs": {
      "command": "npx",
      "args": ["-y", "@dcsplatform/mcp-server"],
      "env": { "DCS_API_KEY": "dcs_••••" }
    }
  }
}
🟠

Cursor

Edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "dcs": {
      "command": "npx",
      "args": ["-y", "@dcsplatform/mcp-server"],
      "env": { "DCS_API_KEY": "dcs_••••" }
    }
  }
}
🔵

Windsurf

Same MCP config schema. Edit Windsurf config in Settings → MCP.

{
  "mcpServers": {
    "dcs": {
      "command": "npx",
      "args": ["-y", "@dcsplatform/mcp-server"],
      "env": { "DCS_API_KEY": "dcs_••••" }
    }
  }
}
Supported clients

Works wherever MCP works.

🟣

Claude Desktop

Anthropic's native client. Tested on macOS + Windows.

🟠

Cursor

AI-first code editor. Tools appear in the chat sidebar.

🔵

Windsurf

Codeium's editor. Same MCP config, same agents.

🟢

Custom MCP host

Anything implementing the MCP spec works. Build your own host.

Tools exposed

641 agents grouped into 5 namespaces.

MCP exposes agents as tools the LLM can call. We group them so the LLM doesn't have to choose between 641 — it picks a namespace, then a specific agent.

🏢

industry.*

590 industry agents across 28 verticals — restaurant, salon, clinic, real estate, retail, logistics, etc.

📄

page.*

48 page-builder agents — hero, pricing, FAQ, testimonials, footer, etc.

🔧

pipeline.*

37 pipeline agents — QA, A11y, Lighthouse, SEO, code review, security audit.

🤝

shared.*

9 cross-cutting agents — translate, summarize, classify, format, schema-extract.

orchestrator

1 meta-agent that delegates among the other 640. Most use cases start here.

🔐

system.*

System tools — list agents, get receipt, verify receipt CID, get pricing for an agent.

Pricing

Free MCP package. Pay only for the agents you actually invoke.

npm package

$0 · free forever

The MCP server itself is free + open source.

Agent calls

Pay-per-call · per Tier 0/1/2

Tier 0 (open-weight on DCS H200): $0.04/M output tokens. Tier 1 (frontier passthrough): vendor cost. Tier 2 (your own weights): infrastructure only.

Free tier

10K calls/mo · no card

Sign up, get an API key, do 10K free calls/month forever.

FAQ

Common questions.

What's MCP?
Model Context Protocol — Anthropic's open standard for AI clients to discover and call external tools. Like USB for AI: plug a tool in, it just appears in your client. Spec at modelcontextprotocol.io.
Is the source open?
Yes — MIT-licensed on github.com/DCS-Platform/mcp-server. Read it, fork it, audit it. We don't get telemetry from your runs.
Does this work offline?
No — the MCP server is a local process but agent execution happens on DCS Compute. Need offline? Use Sovereign tier with your own inference endpoint.
How do receipts work in MCP?
Every tool call → agent call → response includes a receipt_cid in the result metadata. MCP clients display this in their tool result panel. Click through to verify on the public verifier.
Can I write my own MCP server that uses DCS agents?
Absolutely — we publish the agent REST API at api.dcsai.ai. Wrap it in whatever MCP server pattern fits your needs. Our server is just one reference implementation.

One install. 641 agents in your AI client.

30 seconds to "I'm using DCS agents from inside Claude Desktop right now."

Install → REST API instead