Install the MCP server

Nullary speaks the Model Context Protocol, so it drops into any MCP client. Point the client at https://mcp.nullary.ai/mcp — the free tier needs no API key.

One-click install:

Add to CursorAdd to VS Code

Cursor and VS Code open with the server pre-filled (approve the prompt, then restart Cursor if it doesn't appear). Claude Code installs via the copied command. Manual config for each client is below.

Claude Desktop

Edit claude_desktop_config.json directly — on macOS, Settings → Developer → Edit Config opens it for you. Add the server, then restart Claude Desktop:

json
{
  "mcpServers": {
    "nullary": {
      "url": "https://mcp.nullary.ai/mcp"
    }
  }
}

Claude.ai (web)

The web app has no JSON config — you just add the URL. Open Settings → Connectors → Add custom connector, paste https://mcp.nullary.ai/mcp as the remote MCP server URL, and click Add. No API key needed on the free tier.

Cursor

Add to ~/.cursor/mcp.json (or Settings → MCP → Add Server):

json
{
  "mcpServers": {
    "nullary": { "url": "https://mcp.nullary.ai/mcp" }
  }
}

Restart Cursor after adding the server — whether via the snippet above or the “Add to Cursor” button, Cursor doesn't always load a new MCP server until you reopen it. After restarting, it appears under Settings → MCP.

Claude Code

bash
claude mcp add --transport http nullary https://mcp.nullary.ai/mcp

VS Code, Windsurf, Cline & others

Any client that supports remote (HTTP/SSE) MCP servers works — add a server entry pointing at https://mcp.nullary.ai/mcp. The configuration key differs slightly per client (mcp.servers, mcpServers, etc.), but the URL is the same.

What you get

Once connected, your assistant can call tools such as:

  • search_target_history — every failed approach against a target, across all modalities (also per-modality search_* tools and search_indication_history).
  • get_compound — every negative finding for a molecule across sources, with structures and clinical phase.
  • get_finding_provenance — a single finding with full provenance.
  • get_coverage — per-modality and per-source coverage counts.