How to connect Fine Structure to any AI tool that supports MCP

If your tool isn't on the list, use this template: an endpoint, an Authorization header, and an MCP token from Fine Structure.

Most modern AI tools connect to MCP using the same basic idea: you add a server named fine-structure, give it an address, and add an Authorization token. The only difference is the name of the field in the configuration file: some tools use url, some use serverUrl, and some use a special structure like context_servers.

MCP address: https://finestructure.ai/api/mcp is the fixed address of the Fine Structure server.

Token: The header should be Authorization with the value Bearer fse2_mcp_<token>.

Permissions: You choose permissions in Fine Structure: read only, build, publish, secrets, security, or Agency Mode.

Verification: After connecting, ask the agent to call get_platform_guide. If it works, the connection is fine.

The universal way

Create a token, copy the address and the header, and paste them into your tool's MCP screen.

What to look for in any AI tool

• MCP settings
• Model Context Protocol
• Add custom server
• Remote MCP server
• View raw config
• mcp.json or mcp_config.json

Tool | What you usually need to configure

Claude Code | Remote HTTP server with url and an Authorization header.

Codex | config.toml with mcp_servers and a url field.

Google Antigravity | mcp_config.json with serverUrl.

Cursor | .cursor/mcp.json or ~/.cursor/mcp.json.

Windsurf / Cascade | mcp_config.json with serverUrl or url.

VS Code / GitHub Copilot | servers inside mcp.json, ideally with a secret input for the token.

Zed | context_servers inside settings with url and headers.

Cline / Roo Code / Continue / OpenCode / Claude Desktop | Usually the same idea: mcpServers, a server address, and an Authorization header. If there's no support for a remote server, use the MCP bridge that the tool recommends.

{
  "mcpServers": {
    "fine-structure": {
      "url": "https://finestructure.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer fse2_mcp_<token>"
      }
    }
  }
}

{
  "mcpServers": {
    "fine-structure": {
      "serverUrl": "https://finestructure.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer fse2_mcp_<token>"
      }
    }
  }
}

{
  "context_servers": {
    "fine-structure": {
      "url": "https://finestructure.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer fse2_mcp_<token>"
      }
    }
  }
}

What to tell the agent after connecting

• Ask the agent: "Use the Fine Structure MCP and call get_platform_guide first."
• If you're working on an existing app, also give the app name or app_id.
• Before publishing or making a major change, ask the agent to run validate_app and inspect_preview.
• If the connection fails, create a new token at /mcp and revoke the old one so you don't leave access open.

A simple safety rule: Never paste an MCP token into a screenshot, a public issue, a GitHub commit, or a client document. If a token is exposed, revoke it on the /mcp page and create a new one.