MCP Server

Your AI tools can call
Neureus directly

Neureus is a Streamable HTTP MCP server (spec 2024-11-05). Connect Claude Desktop, Cursor, Zed, or any MCP-compatible AI assistant — they can call Neureus tools without you writing glue code.

Get API key MCP vs REST explained
MCP endpoint https://app.neureus.ai/mcp Streamable HTTP · JSON-RPC 2.0 · spec 2024-11-05

Connect in 2 minutes

Claude Desktop

Add to ~/.config/claude/claude_desktop_config.json

{
  "mcpServers": {
    "neureus": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-http",
        "https://app.neureus.ai/mcp"
      ],
      "env": {
        "AUTHORIZATION": "Bearer nr_your_api_key"
      }
    }
  }
}
Cursor / Zed / any HTTP MCP client

Connect directly to the Streamable HTTP endpoint:

// In your MCP client config:
{
  "type": "http",
  "url": "https://app.neureus.ai/mcp",
  "headers": {
    "Authorization": "Bearer nr_your_api_key"
  }
}

// Test with curl:
curl -X POST https://app.neureus.ai/mcp \
  -H "Authorization: Bearer nr_your_key" \
  -H "Content-Type: application/json" \
  -d '{ "jsonrpc": "2.0", "method": "tools/list", "id": 1 }'

9 tools, ready to use

Once connected, your AI assistant can call any of these tools by name. No additional setup required.

neureus_chat
Generate text with any model. Supports streaming, tool use, and system prompts.
neureus_rag_query
Semantic search across your tenant's document index. Returns top-K chunks with source metadata.
neureus_agent_run
Execute a Neureus agent by ID. Returns the final answer and step trace.
neureus_composite_execute
Run a composite AI pattern (ensemble, routing, critique, etc.) on a prompt.
neureus_models
List available models with pricing, context window, and provider info.
neureus_health
Check Neureus API health. Returns status, uptime, and active model providers.
neureus_org_create
Create a new organization grouping multiple tenants.
neureus_org_get
Retrieve organization details and member tenants.
neureus_org_add_tenant
Add a tenant to an existing organization.

What this looks like in practice

After connecting Neureus to Claude Desktop, you can ask Claude to use your Neureus tenant's data directly.

You → Claude
"Search my documents for anything about the Q2 pricing strategy, then generate a one-page summary in a professional tone."
Claude calls Neureus MCP tools ↓
neureus_rag_query({ query: "Q2 pricing strategy", topK: 10 }) → 10 relevant chunks from your docs
neureus_chat({ model: "claude-sonnet-4-6", messages: ["...chunks...", "write a one-page summary"] }) → professional summary
Claude → You
"Based on your Q2 documents, here's the pricing strategy summary: [polished one-page summary grounded in your actual docs]"

Protocol details

Transport
Streamable HTTP (not SSE-only)
Spec version
2024-11-05
Auth
Bearer token (same API key as REST)
Methods
initialize, tools/list, tools/call
HTTP status
200 for valid RPC (including errors), 202 for notifications, 400/503 for transport failures
Tenant resolution
From Bearer token → nr_api_keys D1 lookup. Never from request body.

Let your AI tools use Neureus.

Connect Claude, Cursor, or Zed to your Neureus tenant. All plans include MCP access.

Get API key MCP vs REST guide API reference