Connect AI clients to governed Contentrain projects
Use Studio's remote MCP Connector to let supported AI clients read and update structured project content through OAuth-scoped, Git-backed operations.
$ codex mcp login contentrain
✓ OAuth authorization approved
workspace: Product
project: web-app
scopes: metadata · read · write
✓ One connection · one project
# content writes follow the Git workflow
Overview
Authorize one AI client connection for one project
Add your Studio deployment's /api/mcp/remote endpoint to the AI client. Studio opens in the browser, you sign in, select a workspace and project, review the requested scopes, and approve the connection. Connect again when the client needs a different project.
- Automatic OAuth server discovery from the remote endpoint
- Workspace and project chosen during consent
- One connection is bound to one Contentrain project
Setup
Use the setup path documented for your client
Studio documents remote MCP setup for Claude, Claude Code, ChatGPT developer mode, Codex, and MCP Inspector. Each client uses the same endpoint and browser consent flow, even though its settings or configuration file differs.
- Claude custom connector and Claude Code HTTP transport
- ChatGPT developer mode remote MCP server
- Codex MCP server configuration and login command
- MCP Inspector for discovery and debugging
Tools
Grant only the project and content scopes the client needs
OAuth scopes separate project metadata, content reads, content writes, and offline access. Eligible deployments can also advertise media read and write scopes. Tool calls use the canonical contentrain_* MCP tool family and remain subject to tool availability.
- contentrain_status and contentrain_describe for project context
- contentrain_content_list and contentrain_content_save for governed content work
- No remote contentrain_merge, contentrain_submit, or branch lifecycle access
- Media tools appear only with the required stack, plan, and project settings
Availability
Confirm deployment, plan, quota, and media requirements
The remote endpoint requires managed authentication with the PostgreSQL database provider. It is plan-gated, shares the workspace MCP call quota with API keys, and only exposes media scopes when the deployment, storage, plan, and project CDN settings satisfy the media capability checks.
- Community, Starter, Pro, or Enterprise plan; not the Free shell
- Managed authentication plus PostgreSQL database provider
- Shared monthly MCP-call quota and per-connection rate limits
- Additional storage and project gates for media operations
Privacy
Keep access bounded, revocable, and auditable
Access tokens expire after one hour and refresh tokens can keep an approved client connected for up to 30 days. Disconnecting an app revokes its token family. Content writes remain Git-backed, while the privacy addendum explains data routing, encryption, retention, sub-processors, and deletion.
- Scoped OAuth grants for one selected project
- Immediate connection revocation from Connected Apps
- Git commits and project workflow for content writes
- Connector-specific data handling documented in the privacy addendum
Common questions
What is the Contentrain MCP Connector?
It is Studio's OAuth-authenticated remote MCP endpoint. A supported AI client connects to one chosen project and calls bounded content tools without receiving direct repository credentials.
Which AI clients have documented setup paths?
The Studio codebase documents Claude, Claude Code, ChatGPT developer mode, Codex, and MCP Inspector. Other clients can connect when they support remote HTTP MCP and the required OAuth flow.
Is the connector the same as the local MCP package?
No. The MIT @contentrain/mcp package can run locally over stdio or HTTP. The Contentrain MCP Connector is Studio's shared remote endpoint with OAuth, project selection, scoped access, quotas, and deployment gates.
Where is the remote MCP Connector available?
It requires Studio's managed authentication and PostgreSQL provider pair plus an eligible Community, Starter, Pro, or Enterprise plan. The Free workspace shell is not eligible.
Can the connector merge or reject review branches?
No. Studio-owned branch lifecycle tools are not exposed over remote MCP. Content writes land on cr/* branches and follow the project's configured auto-merge or review workflow.
Can I revoke a connection?
Yes. Studio's Connected Apps settings let authorized users disconnect an app. Revocation invalidates its token family and the client must authorize again before making another call.
Next paths
Continue through the content system
Start local. Scale to Studio.
Build a governed content layer before content becomes product debt.
Start locally with the MIT packages. Add Studio when teams need review, roles, structured editing, managed delivery, or deployment control.