Real user, real permissions
Every agent call authenticates with a JWT for an actual user and runs as that person. There is no superuser agent.
AI and agents
Let agents work on your content. Keep the keys.
Formscaler ships an authenticated MCP server that runs on the same services and permissions as your admin. Agents draft. People publish.
The danger of AI in your CMS isn't the model. It's the unguarded database behind it.
Built for trust
Schema-aware tools
Agents discover collections and fields they may read, with JSON Schema for each, then list, read, create and update records.
Drafts, not publishes
In v1 agents create and edit drafts. Publishing is not exposed over MCP, so a person always signs off.
No silent overwrites
Updates and deletes require the current lock version, so an agent cannot clobber an editor's work.
Audited, rate limited
Denied calls are written to the audit log. Each user is limited to 60 calls per minute.
Standards-based
JSON-RPC 2.0 over HTTP following the MCP specification. Works with MCP clients out of the box.
Governance
One permission system for people and agents.
Give an agent its own role with read access to products and draft access to articles. Revoke it in the same screen you use for everyone else.
Audit
See exactly what the agent tried.
Every created record and every denied call is recorded. Enterprise adds per-agent scopes, tool-call audit streams and tighter rate limits.
What teams will do with it
Content operations
Draft changelogs, translate entries, fill product metadata, and queue everything for review.
Support and knowledge
Agents read approved articles to answer questions without seeing restricted fields.
Data hygiene
Find incomplete records, propose fixes as drafts, and let editors approve in bulk.
Give your agents guardrails.
Start with the MCP server in the free core.