Ailimu.

MCP connector for FileMaker Server

Ask your FileMaker data anything.

Ailimu turns a FileMaker solution into clear tools that Claude, or any MCP client, can call. Skip the layouts and exports and just ask. It runs on your server, starts read-only, and sets up in about ten minutes.

Read the install guide See the setup
$ docker pull ailimu/ailimu
Read-only until you say otherwise Chosen field by field Your privilege sets still apply Runs inside your own network
Assistantailimu · harbour-finch
Which trade customers have overdue invoices?
Connected via Ailimu. Harbour & Finch, 24 tools
find_invoices(payment_status: "Overdue")→ 11 invoices
get_customers × 7→ segment, contact, credit limit

Five trade customers are overdue, owing £28,940 between them. Greystone Hardware owes the most.

Greystone Hardware3 invoices£9,214
Copperfield Kitchens2 invoices£7,480
Rowan Catering2 invoices£5,105
Kestrel Marine1 invoice£4,012
Saltmarsh Brewing1 invoice£3,129
Try a question

A scripted example with an invented business. Every name and figure is made up.

What the agent gets

Your schema becomes tools shaped around real questions.

Hand an agent a raw connection and it has to guess which of four copies of a table is the real one, what zz_PrintFlag is for, and whether an empty column matters. Ailimu settles those questions during setup, with you checking its work.

Orders
18 fields, and one of four occurrences of the same table
  • OrderIDkey
  • _CustomerIDlink
  • OrderDatedate
  • Status5 values
  • OldRefNumbernever filled in
  • zz_PrintFlagscaffolding
  • zs_TotalValuesummary
  • zs_OrderCountsummary
  • ModificationTimestampautomatic
  • Totalamount

and 8 more

find_orders

Search orders (18,430 rows). Dated transactions, linked to customers, sales reps and warehouses. Order number is SO- followed by five digits. Dates accept >=2026-01-01 or 2026-01-01..2026-03-31.

order_datedate, ranges and comparisons
statusDraft, Confirmed, Shipped, Invoiced or Cancelled
customer_idresolves with get_customers
totalnumber, ranges and comparisons

Descriptions come from the field comments already in your file. Every table also gets get_ and describe_.

Why Ailimu

Set up in minutes, and stay in control.

Reads your schema for you

Ailimu reads your real schema: types, keys, relationships, row counts and field comments. It sorts tables into core records, transactions, lookups and logs, and suggests where to start. Nobody writes a tool definition.

Safe by default

  • Read-only until you switch it off. While it is on, no update or script tool exists.
  • You choose each field: returned, searchable, or invisible.
  • It signs in as an account you create, so your privilege sets stay in charge.
  • Every write and script call is logged.

Built for real FileMaker files

  • Tables from the data file, scripts from the interface file, on one endpoint.
  • Duplicate table occurrences collapse to one set of tools.
  • Several files at once, each with its own address and token.
  • Re-read the schema and every choice you made is kept.

The setup wizard

Four steps in your browser, from file to live tools.

Shown here with a demo file for an invented wholesaler, Harbour & Finch.

Connect

Point it at your server and pick the file. Tables can come from one file and scripts from another.

Connection details for the Harbour_Data file, with scripts run from Harbour_UI.

Self-hosted

See exactly what leaves your network.

Ailimu is one container you run on your own server, next to FileMaker. Your password, your settings and the audit log live there, and nothing is copied out. An agent can only answer by reading data, so the rows a tool returns go to the model when it asks, and nothing else does.

YOUR NETWORK your server, your firewall, your rules Your admin setup wizard in a browser chooses what is exposed FileMaker Server your solution, unchanged privilege sets still apply Ailimu. one Docker container password stays in its .env config and audit log on its volume reads live exposed fields only AI agent Claude or any MCP client TOOL CALL SELECTED ROWS MCP over HTTPS a separate token per file NEVER LEAVES FileMaker password Unexposed fields The database itself Config and audit log
Every call reads FileMaker live. Nothing is synced, mirrored or indexed, so switching Ailimu off leaves no copy of your data behind.

Never leaves your server

  • Your FileMaker credentials. Read from the environment, never written to config, never sent to a browser.
  • Anything you did not expose. It is not in any tool, so it cannot be asked for.
  • The database as a whole. Nothing is synced, mirrored or indexed. Every call reads FileMaker live.
  • Container fields. Excluded from every tool.

Sent to the model when a tool is called

  • The rows that tool returned, limited to the fields you marked readable.
  • Tool names and descriptions, so the agent knows what it may ask for.

If your data may not reach a third-party model at all, no connector of this kind is right for you. That is a decision about the model, and it is worth making before you install anything. The full answer is in the docs.

Get started

Ten minutes to your first answer.

$ docker pull ailimu/ailimu
Install guide →
  1. Create a FileMaker account for AilimuA restricted privilege set with fmodata, plus fmrest if you want scripts.
  2. Run the containerOne image, one volume, two environment variables. It prints its setup link on first start.
  3. Pick tables and fieldsThe wizard suggests a starting set. Adjust it and try each tool against live data.
  4. Publish and connectPaste one command into Claude Code, or one block into Claude Desktop.