The 24 tools
You never call these yourself. You ask a question, and the AI picks the tools that answer it. This page describes what a connected AI can reach: all of it read-only, all of it scoped to one client. We describe the tools by the jobs they do rather than as a tool-by-tool reference; the AI reads the full technical descriptions when it connects, so you never need them.
Some of these draw on server logs and only have something to say once you've uploaded them.
Get oriented
Tools that give the AI the lay of the land: what the solution is, what files it spans, what there is to ask about. Usually the AI's first move in a fresh conversation.
"Give me the lay of the land."
Find things
Search across every kind of object at once (typo-tolerant), search inside script logic, query every script step in the solution, track a variable, and answer superlatives like which table is biggest or which script is longest.
"Every place any script deletes records."
Look closely
Detail views for any single object (a field, a script, a layout, a relationship, a privilege set, and the rest), one-level-deeper dives into everything that touches it, raw script listings, and stepwise script audits with an eye for error-handling gaps and risky steps.
"What touches this field?"
Production evidence
The log-backed tools: errors that actually fired, the slowest and costliest operations from real timings, a timeline of everything happening around a moment on the server, schedule behavior, and an overall server health report.
"Build a timeline around 2:30pm on the 14th."
Check and clean
Unused-element reports with the evidence for each verdict, security audits across accounts and privilege sets, and web-readiness checks that test whether a table can safely go over the Data API or OData with a given privilege set, backed by FM Fathom's verified integration rules. Reference lookups mean answers about script steps and functions come from documentation, not memory.
"Is this table safe to expose over OData?"
Compose
The generating tools; none of them write to your file. Draft scripts and field definitions against your real schema, or turn an existing script into a snippet for moving logic between files. Output is paste-ready XML you review yourself.
"Draft the script that archives closed invoices."
Related
- Prompt recipes: these tools in real workflows
- Connect Claude if you haven't yet