# Connect over MCP

> Add DECA to Claude, Cursor, VS Code or your own agent through the Model Context Protocol.

DECA is a remote MCP server. It speaks streamable HTTP, answers every request with JSON and keeps no session, so any client that supports remote servers can use it.

Server URL: `https://app.decalabs.ai/mcp`

Claude Code: `claude mcp add --transport http deca https://app.decalabs.ai/mcp`

| | |
|---|---|
| **Server URL** | `https://app.decalabs.ai/mcp` |
| **Transport** | Streamable HTTP, JSON responses, no server-sent stream |
| **Authentication** | None needed. An optional API key ties calls to your workspace; see [Authentication](https://app.decalabs.ai/docs/reference/authentication). |
| **Protocol versions** | `2025-11-25`, `2025-06-18`, `2025-03-26`, `2024-11-05` |
| **Tools** | `get_signals`, `get_ideas`, `get_my_ideas`, `quote_test`, `start_test`, `get_test`, `list_my_tests`, `cancel_test` |
| **Registry** | `ai.decalabs/deca` in the [official MCP Registry](https://registry.modelcontextprotocol.io/v0.1/servers?search=ai.decalabs/deca) |

## Clients

```bash title="Claude Code"
claude mcp add --transport http deca https://app.decalabs.ai/mcp
```

```json title="Claude Desktop"
{
  "mcpServers": {
    "deca": { "type": "http", "url": "https://app.decalabs.ai/mcp" }
  }
}
```

```json title="Cursor"
{
  "mcpServers": {
    "deca": { "url": "https://app.decalabs.ai/mcp" }
  }
}
```

```json title="VS Code"
{
  "servers": {
    "deca": { "type": "http", "url": "https://app.decalabs.ai/mcp" }
  }
}
```

In Claude.ai, add it under **Settings → Connectors → Add custom connector** with the server URL.

## From your own agent

Any MCP client library works. With the official TypeScript SDK:

```ts title="TypeScript"
import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";

const client = new Client({ name: "my-agent", version: "1.0.0" });
await client.connect(new StreamableHTTPClientTransport(new URL("https://app.decalabs.ai/mcp")));

const { tools } = await client.listTools();
const result = await client.callTool({ name: "get_ideas", arguments: { query: "camera", limit: 3 } });
console.log(result.structuredContent);
```

```python title="Python"
import asyncio
from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async def main():
    async with streamablehttp_client("https://app.decalabs.ai/mcp") as (read, write, _):
        async with ClientSession(read, write) as session:
            await session.initialize()
            result = await session.call_tool("get_ideas", {"query": "camera", "limit": 3})
            print(result.structuredContent)

asyncio.run(main())
```

```bash title="JSON-RPC"
curl -X POST https://app.decalabs.ai/mcp \
  -H "content-type: application/json" \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": {"name": "get_signals", "arguments": {"heat": "rising", "limit": 2}}}'
```

## What a tool returns

Every tool result carries the same object twice: as `structuredContent` for clients that read it, and as JSON text in `content` for those that do not. A failed call sets `isError: true` and its object is an [error](https://app.decalabs.ai/docs/reference/errors).

```json
{
  "content": [{ "type": "text", "text": "{\"day\":\"2026-11-18\",\"count\":2,\"ideas\":[…]}" }],
  "structuredContent": { "day": "2026-11-18", "count": 2, "ideas": ["…"] },
  "isError": false
}
```

> **Tip** · The server's `instructions`, sent on `initialize`, tell the model what DECA is and when to use each tool. Most agents need nothing more than the server URL.

Source: https://app.decalabs.ai/docs/mcp
