Singular MCP

Installation

No package install is published in the registry. Use the remote MCP endpoint https://mcp.singular.net/mcp-server/mcp.

Authentication

Access uses Singular login/OAuth. Supported clients initiate a Singular sign-in flow and then operate with the permissions of the connected Singular account.

Supported clients

ChatGPT, Claude and Claude Code (paid version), Cursor, Visual Studio Code Copilot, and other MCP-capable clients.

Configuration

Remote transports are available at https://mcp.singular.net/mcp-server/mcp.

• ChatGPT: open Singular's GPT page, sign in, confirm the connection, and optionally set the MCP domain to always allow.
• Claude / Claude Code: add a custom connector with the integration URL above, then connect and sign in to Singular.
• Cursor: add a custom MCP server using HTTP transport with the same URL, then complete the Singular OAuth login flow.
• Gemini CLI: add the server to ~/.gemini/settings.json with httpUrl set to https://mcp.singular.net/mcp-server/mcp, OAuth authorization and token URLs under Singular's OAuth server, clientId Gemini, and scope mcp:read; then run gemini and /mcp auth singular-mcp.
• VS Code Copilot: requires VS Code 1.101 or higher, an active GitHub Copilot subscription, and MCP servers enabled in the workspace.

Capabilities

The MCP returns the data available in Singular's main report, including aggregated tracker data from SDK or S2S and cost data from networks. It also exposes get_report and the field-management tools get_supported_fields and update_supported_fields.

Limitations

Only aggregated data is available; user-level data is not exposed. SKAN data is coming soon. LLM clients impose response-size limits, with ChatGPT limited to about 50k characters per query and Claude to about 100k characters. ChatGPT workspaces may require third-party GPTs to be enabled by an admin. Singular states that context data is not used for model training and is encrypted in transit and at rest.