Intelligent Coding

CODE mode is usually reliable when context and prompts are clear. This page shows how to get better results and how Genie fits with IDE agents through Local MCP.

What improves reliability

Genie uses your attached context (scripts, selections, metadata).
Preflight questions reduce ambiguity before generation.
Output is aligned to FileMaker step/function vocabulary.
Larger or more complex CODE requests preserve essential FileMaker step patterns more consistently, which reduces incomplete first-pass output and follow-up repair cycles.
Sources may be shown so you can verify references.

Local MCP brings agentic development to FileMaker

The Local MCP Bridge turns Genie into the FileMaker-aware workbench for agents outside the app, such as Codex, Cursor, or another local MCP-aware IDE agent. It gives those agents grounded FileMaker context instead of asking them to guess from source files or loose prose.

Through MCP, an agent can:

read Genie resources for metadata, selected metadata, Script Schema, Script Catalog, and active script context
use genie_files for active/open FileMaker file context
use genie_scripts for script context, content, schema, catalog, steps, functions, validation, and staging
use genie_metadata for layout, table, field, relationship, value-list, and metadata-selection work
use genie_evaluate to check FileMaker calculations before staging
use genie_query for read-only SELECT preview or run workflows outside Workbench
ask genie_get_capabilities before choosing a route
stage one supported script or DATA request into Workbench for user review
inspect staged Workbench items and job status

The bridge normally does not let the external agent silently generate, apply, or mutate FileMaker during the MCP tool call. It is the agentic workbench boundary: the agent can gather context and stage work, then Genie owns review, CODE safety, DATA confirmations, and apply behavior. The beta Automatically create scripts in FileMaker setting is an explicit opt-in exception for trusted local clients: supported successful script stages can immediately run Genie’s Create Script path, and existing scripts are never replaced.

MCP-to-Workbench workflow

For CODE and DATA work, the intended external-agent flow is:

Call genie_get_capabilities.
For script work, use genie_scripts action:get_schema, targeted catalog/step/function reads, and genie://context/scripts before edits.
Read the relevant metadata/context resources or use genie_metadata.
Use genie_query only for read-only query work outside Workbench.
Prepare any needed metadata selection with genie_metadata action:parse_selection.
Stage exactly one supported item with genie_scripts or genie_stage_data.
Review and run the item from Workbench inside Genie.

Workbench staging is separated by intent. Script creation and active-script edits go through genie_scripts with Script Schema JSON. Reviewed SQL data/schema work goes through genie_stage_data. ASK/planning remains in the IDE or normal chat, and QUERY remains a read-only MCP tool path.

Calculation Quick AI

Calculation Quick AI is a separate native overlay inside FileMaker’s Specify Calculation dialog. It is for drafting one calculation expression from the current dialog context.

It does not use the CODE script workflow, QuickCode authority, script apply, or chat-mode apply actions. The calculation pipeline sends a tiny calculation-focused context to the Coding provider, validates the structured calculation draft, rejects script-looking output, checks known fields and functions where possible, and inserts only draft expression text after dialog drift checks pass.

Use it when you are already in a calculation dialog and want a fast draft expression. Use CODE when you need FileMaker scripts. Use MCP Workbench when an IDE agent is handing Genie a reviewed CODE or DATA task.

What you can do to help

Attach only relevant script and schema context.
Use explicit names for tables, fields, and scripts.
Answer preflight questions with concrete choices.
Preview before applying high-impact changes.

When results are weaker

Context is too broad or stale.
Prompt uses ambiguous names.
Required metadata was not selected.

Why larger requests are more stable now

Genie is better at carrying forward essential structural patterns in CODE mode, especially when prompts involve common control flow or save/commit behavior.

Fewer missing structural steps in larger outputs.
More stable first-pass results before repair is needed.
Better continuity when your request references branching, script exits, field updates, or record-commit behavior.