ATLAS MCP Integration

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets agentic AI frameworks communicate with external data sources in a structured, tool-like way. Rather than scraping websites or writing custom API wrappers, your LLM agent can call ATLAS tools directly — searching datasets, inspecting schemas, and exploring facets — all through a single authenticated connection.

ATLAS exposes its catalogue via MCP so that developers, researchers, and enterprises can integrate high-quality SEA language data into their AI pipelines without leaving their existing toolchain.

How ATLAS Exposes MCP

Once connected, your agentic framework can access these MCP tools:

Dataset Search

Search datasets by keyword, language, modality, task, domain, or licence.

Catalog Facets

Explore the top languages, tasks, domains, and modalities across the catalogue.

Dataset Health

Audit a dataset's production readiness, licensing, and metadata completeness.

Config Listing

List all available subsets and configurations for a given dataset.

Schema Inspection

Retrieve features, splits, and download size for a specific dataset config.

Quick Start

Connect Claude Desktop or any MCP-compatible client to ATLAS using the configuration below. Replace YOUR_API_KEY with the key from your profile.

Claude Desktop config

json

{
  "mcpServers": {
    "atlas": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://dj3jjj4wljyni.cloudfront.net/mcp",
        "--header",
        "Authorization: Bearer YOUR_API_KEY"
      ]
    }
  }
}

Claude Desktop does not natively support Streamable HTTP. Use the proxy command shown above instead.

Remote MCP clients (Streamable HTTP)

text

// Endpoint
https://dj3jjj4wljyni.cloudfront.net/mcp

// Transport
streamable-http

// Required header
Authorization: Bearer YOUR_API_KEY

// Fallback (if custom headers unsupported)
?api_key=YOUR_API_KEY

For clients that support Streamable HTTP transport, connect using the endpoint and Bearer token header shown above. If your client does not support custom headers, you may fall back to passing the key as a query parameter — note that query parameters are visible in server logs.

Authentication

Every request to the ATLAS MCP server must include your API key in the Authorization header as a Bearer token. Your key grants read-only access to the public catalogue — no data can be modified or deleted through the MCP interface.

Security & Limits

API key required on every request via Authorization: Bearer header.

The API can only read data — nothing can be modified or deleted.

All queries are sanitised to prevent injection attacks.

Results are capped and only relevant fields are returned.

Troubleshooting

Connection Refused

Your MCP client must support Streamable HTTP transport. Claude Desktop does not support this natively — use the npx mcp-remote proxy command instead.

401 Unauthorized

Your API key is missing or incorrect. Ensure the request includes the Authorization: Bearer YOUR_API_KEY header. You can find or regenerate your key in your profile.

Empty or No Results

The search was too narrow. Start with only 1–2 filters (e.g. language + task) and broaden from there. Use the Catalog Facets tool to discover valid values.

Dataset Config Not Found

Config names are case-sensitive. First call list_dataset_configs to see exact config names, then pass the correct name to get_config_schema.