MCP and API Credentials

Dawn can act as an MCP server for your own account. That lets you use your existing Dawn workspaces, agents, and private MCP threads from tools such as Codex, Cursor, and GitHub Copilot.

What this page is for

• Use this page when you want an external MCP client to call Dawn.
• Create a personal API credential from Account settings → API credentials.
• Point the client at your Dawn API origin plus /mcp.
• Authenticate with a normal bearer token.

If you want Dawn itself inside VS Code, use VS Code Extension. This page is for external MCP clients, not the first-party Dawn editor extension.

This is separate from Link Existing OpenAI ChatGPT Account. That page is about using OpenAI models through a linked OpenAI ChatGPT account inside Dawn. This page is about using Dawn as an MCP server from other tools.

It is also separate from MCP Gateway. MCP Gateway is the opposite direction: Dawn connects to hosted MCP servers as workspace integrations and makes selected capabilities available to agents.

Create the credential

1. Open Dawn Web.
2. Open Account settings.
3. Go to API credentials.
4. Create a token with a clear name such as Codex, Cursor, or GitHub Copilot.
5. Copy the raw token immediately. Dawn only shows it once at creation time.
6. Revoke the token later from the same screen if you no longer need that client.

Endpoint and authentication

• Production endpoint: https://api.dawnhq.ai/mcp
• Authentication header: Authorization: Bearer dmcp_...
• The token acts as the user who created it. The client only sees the workspaces and agents that user can already access.
• Revoking the token immediately cuts off that MCP client.

What Dawn MCP can do

• List your accessible workspaces.
• List the enabled agents in a workspace.
• Create and continue private MCP threads.
• Send messages to an agent and read the replies in that thread.

MCP threads are personal to the authenticated user. They are not the shared workspace conversation view. Use shared channels when the team should see the thread, and MCP when you want a private editor-side workflow.

Codex example

[mcp_servers.dawn]
url = "https://api.dawnhq.ai/mcp"
bearer_token_env_var = "DAWN_MCP_TOKEN"

Start with a tool-level check, not a knowledge question. First ask the client to list your Dawn workspaces. Then ask it to list the agents in one workspace. Only after that should you test a real agent message. If the workspace list fails, the endpoint or token is wrong.

Cursor example

{
  "mcpServers": {
    "dawn": {
      "url": "https://api.dawnhq.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${env:DAWN_MCP_TOKEN}"
      }
    }
  }
}

Restart Cursor after setting the environment variable so the header is available when Cursor opens the MCP connection.

GitHub Copilot example

{
  "servers": {
    "dawn": {
      "type": "http",
      "url": "https://api.dawnhq.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${input:dawn-mcp-token}"
      }
    }
  },
  "inputs": [
    {
      "id": "dawn-mcp-token",
      "type": "promptString",
      "description": "Dawn MCP token",
      "password": true
    }
  ]
}

Copilot Agent mode can prompt you for the token and then call the same Dawn MCP endpoint over HTTP.

Troubleshooting

• 401 or authentication failure: recreate the token if the raw value was lost, or revoke and issue a new one if you suspect exposure.
• No workspaces or agents appear: the token only has the access of the Dawn user who created it. Confirm the same account can see those workspaces in Dawn Web.
• The client connects but agent questions are unreliable: first verify the client can list workspaces and agents. That tests MCP access directly without depending on an agent answer.
• MCP tools are missing: check that the URL ends with /mcp and that the client is really sending the bearer token.
• You wanted linked OpenAI ChatGPT account access inside Dawn instead: use Link Existing OpenAI ChatGPT Account, not this page.