Connect your AI, an overview
This is the part of FM Fathom where your solution stops being something you browse and starts being something you can ask. Here's how it works and what to trust about it.
What the connector is
FM Fathom includes an AI connector built on MCP (Model Context Protocol), an open standard that lets AI apps like Claude use outside tools. When you connect an AI client to FM Fathom, that AI gains 24 read-only MCP tools for looking things up in your solution: search it, describe any object, trace a script's callers, check web readiness, rank tables by field count, and so on. The AI decides which tools to call based on what you ask, and answers with specifics from your actual file.
The connector is scoped to one client. If you work with three clients in FM Fathom, each gets its own connection, and a conversation about one can never wander into another.
The read-only promise
Every tool reads. None of them write. The connector cannot change your FileMaker file, your data, or anything in FM Fathom. Two clarifications worth having in writing:
- A few tools compose: they generate script or field definitions as paste-ready snippets, grounded in your real schema. Generating a snippet doesn't touch your file. You review it and you paste it, or you don't.
- The AI sees your solution's structure (schema, scripts, layouts, security objects) and, if you've uploaded logs, production evidence. It never sees your database records, because those aren't in the XML export to begin with. More on this in Your data.
Three ways to ask
- Claude, on the web or in Claude Desktop: add the connector URL from your Connect to AI page, sign in, done. No key to copy, nothing to paste into a config file.
- Claude Code, Codex, and other MCP clients: one command or a small config block, using either the sign-in flow or a static access key.
- The built-in assistant: chat inside the portal itself, no external app needed. Bring your own API key.
What asking looks like
What would be the impact of renaming the field Customers::Email?
Walk me through what "Process Invoice" actually does, step by step.
Is the Invoices table safe to expose over OData with the Sales privilege set?
The prompt recipes page has a dozen of these, organized by job.
Related
- All 24 tools, if you want to know exactly what the AI can reach
- Prompt recipes for real workflows