MCP Server
Connect an AI Agent with MCP
Connect Claude, Cursor, or ChatGPT to Doxa through the remote MCP server and OAuth sign-in
What the connection gives your agent
Doxa runs a remote MCP server that lets an AI agent work with your projects. Once connected, the agent can list projects, read and write pages on branches, build previews, and publish, within the scopes you grant. You connect the MCP client once, sign in with OAuth, and choose what the agent can access.
Supported clients include Claude, Cursor, and ChatGPT.
Before you start
A Doxa account. Sign in to the dashboard with Google or GitHub.
At least one project. If you have not created one yet, follow the generation workflow in the Doxa Platform Overview first.
Access to the settings of the MCP client you want to use.
Connect the agent
Copy the connection snippet
Open the new-project screen in the dashboard and find the MCP connect card. Click the copy button to copy the snippet. The button copies the exact text, including the server address.
Paste the snippet as it is. Do not edit the server address by hand.
Add the remote server to your MCP client
In your client's settings, add a remote MCP server using the address from the snippet. Use the client's own screen for remote or custom MCP servers:
Claude: add a custom connector.
Cursor: add an MCP server in the MCP settings.
ChatGPT: add a custom MCP connector.
The server address is the same in every client.
Complete OAuth sign-in
When the client opens the Doxa sign-in page, sign in with the account you use for the dashboard. The client receives tokens only after you sign in. Access tokens last 1 hour. Refresh tokens last 30 days, or 14 days without use.
Choose scopes and projects on the consent page
The consent page at /oauth/consent asks what the agent may do and which projects it may use.
| Choice | What to decide |
|---|---|
| Scopes | The actions the agent can take. Grant only what the task needs. See MCP Scopes and Approval Requests. |
| Projects | The projects the agent can access. Leave out projects the agent does not need. |
Approve the request to finish connecting the agent.
Check the connection
Ask the agent to list your projects. It calls the list_projects tool, described in Account and projects. The agent should return the projects you selected on the consent page.
Note:
Scopes limit what the agent can do. Writes to main are rejected through MCP, so agents work on branches. Consequential actions can require an approval request, which you review in the dashboard. See Approval requests.
Review or revoke a connected agent
Each approved client appears as a grant under Settings, Connected agents. Use this page to:
See which client is connected and which scopes and projects it has.
Revoke a grant when you no longer want the agent to have access.
To change what an agent can do, revoke its grant and connect again, so the consent page records your new choices.
Troubleshooting
| Symptom | What to check |
|---|---|
| The client cannot connect | Confirm you pasted the full snippet from the MCP connect card without edits. |
| The agent sees no projects | Reconnect and select the projects on the consent page. |
| An action is refused | Check the scopes on the grant under Connected agents. |
| A change is waiting for approval | The action is an approval request. Review it in the dashboard. |
What's next
What each scope allows and how approval requests work.
Tools for listing, reading, creating, and deleting projects.
Tools for reading, searching, and writing pages.
Tools for publishing branches and restoring earlier publications.