MCP Servers

OttoFMS MCP (Model Context Protocol) Servers are a method to expose your FileMaker Applications to AI clients like Claude, Cursor, and Goose. You can define your own MCP servers and tools in OttoFMS and use them from AI clients in a chat like interface. For more information on the MCP servers and tools check out the MCP documentation (opens in a new tab).

OttoFMS MCP Servers are a wrapper for the FileMaker OData API allowing you to expose scripts that you write in your FileMaker Applications as part of an MCP server. This means that you can write any sort of behavior you want from within FileMaker and expose it without any need to use an extra tool or service (other than OttoFMS).

OttoFMS OData MCP Servers are authenticated using Data API Keys. The Data API Key used must be for the FileMaker file the MCP server is connected to and it must have OData privileges.

Getting Started

You'll need a couple things prepared before you can create an MCP Server:

1. A FileMaker Application you want to expose functionality or data from.
2. A Data API Key for the FileMaker Application that has OData privileges. (Both the User account and the Data API Key need to be enabled for OData). Ideally, this key is also enabled for the Data API, as we use the Data API to query the list of scripts to make tool creation easier.

Script Preperation

Scripts for the MCP Server will take in a JSON Object as the Script parameter. You can define the structure of the JSON object when you define the tool parameters, but OttoFMS currently only supports top level keys and string, number, and boolean values. String and number keys can be arrays. This may change in a future version.

Example JSON object:

{
  "firstName": "John",
  "lastName": "Doe",
  "age": 30,
  "isStudent": true
}

Unsupported JSON object:

{
  "type": "Person",
  "param": [
    {
      "name": "John",
      "age": 30
    },
    {
      "name": "Jane",
      "age": 25
    }
  ]
}

MCP Scripts should return either a string or a JSON object. This value will be passed directly back to the AI client as the response to the tool call, so it should include all information you may want to expose to the user. Unless you are using a private LLM, all data sent back from tools will be sent to the LLM, so be mindful of privacy concerns.

Creating an MCP Server

To create an MCP Server, you'll need to navigate to the MCP Servers page in the OttoFMS console.

Click the "Create MCP Server" button in the upper right to create a new MCP Server.

Fill out the form in the modal dialog that pops up. When you have entered everything correctly click the "Create Server" Button.

The MCP Server description will be exposed to the AI client to help it understand which server does what. The slug is used for the server in urls and should be unique.

Adding tools

To add a tool, click the "Add Tool" button in the upper right of the MCP Server page.

The modal that pops up has three tabs, which will let you define the tools name, description, script, parameters, and run a test.

The General tab will let you define the tools name, description, and script as well as the active status. The active status will determine if the tool is exposed to the AI client. The name and description will be used by the AI Client to help it understand which tool does what, so it is important to make them descriptive.

The Input Schema tab will let you define the parameters that the tool will take. Each parameter is a key in the JSON object passed to the tool.

Each parameter has a name and description, as well as a type. The type can be one of the following:

• String
• Number
• Boolean

Strings and Numbers can have the array option toggled on, in which case the tool will take an array of strings or numbers for that key in the JSON object.

Strings can have enum values defined, which will limit the values that can be passed in to a predefined list.

A parameter can also be marked as required, which will require the parameter to be passed in when the tool is called. Optional parameters may be omited from the JSON object passed to the tool.

The Test tab can be used to test the tools functionality (or run the script remotely if you just want to trigger it manually). This will directly call the script via OData so you can see the result given based on the parameters you passed in.

Using the MCP Server with AI clients

Once you have created an MCP Server and added tools, you can use it with AI clients like Claude, Cursor, and Goose.

{
  "mcpServers": {
    "your-custom-name": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@latest",
        "https://[your.server.hostname]/otto/mcp/[your-mcp-server-slug]",
        "--header",
        "Authorization: Bearer [data-api-key]"
      ]
    }
  }
}