Insivia Labs MCP
AI assistant integration and content tooling
Bringing website content into an AI conversation
The Labs MCP server connects compatible AI assistants to the same content workspace. It gives an authorized user tools to read content, create drafts and make specific edits without rebuilding the workflow in every assistant.
In development
Why connect an AI assistant to the CMS?
The challenge
Copying content into a chat loses the page structure and the context of who can edit it. An assistant needs enough information to make a focused change, and the CMS still needs to enforce access and catch conflicting edits.
What can the MCP tools do?
My work
I built tools for reading assigned workspaces, editing structured page sections, creating drafts and managing media, HTML graphics and form schemas. Markdown imports have a preview step, while revision tools make earlier content available for review and recovery.
The server uses individual OAuth identities and the existing CMS permissions. Writes require the current content version, so an assistant has to reread a changed item rather than overwrite someone else's newer work.
Technical details
- Node.js and the official Model Context Protocol SDK, using Streamable HTTP.
- Supabase OAuth with PKCE, scoped user tokens, role checks and row-level access controls.
- Structured content tools, reviewed Markdown imports, validated image uploads and revision recovery.
- No staging, release approval, production publishing or user-management tools. Administrator-only shared site settings affect public headers and footers immediately.
How do edits stay controlled?
Result
Page and library edits stay in development, with staging, review and publication handled in the CMS. Administrator-only shared header and footer settings are a separate action that applies immediately to previews and public pages.
Is the connection ready for everyone?
Status
The server is hosted, with ChatGPT and Claude connection support implemented. Setup and tool availability depend on each client; sign-in, token refresh and editing still need end-to-end verification after the custom-domain change.
Want to talk about the work?
Connect on LinkedIn