Skip to content

AI Tooling (MCP + Skill)

FHIR Frog ships an MCP server and an agent skill so an AI assistant (Claude Code, Claude Desktop, Kiro) can do the day-to-day work of testing FHIR implementations for you: scaffold a Test IG, write and compile TestScripts, bring up a local test server and runner, run tests locally or remotely, and explain the failures.

The two pieces do different jobs:

Piece What it is What it gives the assistant
MCP server (fhir-frog-mcp) A program the assistant talks to over MCP. 24 tools in v0.1 The ability to do things: lint a script, start containers, run tests, read reports
Skill (fhir-frog) A folder of instructions the assistant loads The knowledge of how and in what order, including where the published docs are wrong

You need both for the best results. The tools work without the skill, but the assistant then has to guess the TestScript Shorthand grammar and the run workflow.

Who it is for

  • IG developers who want tests that travel with their implementation guide: scaffold the IG or a separate Test IG, write tests, compile, run, wire the build.
  • System testers who want to point a known set of tests at a server: set up a local server and runner (or use remote ones), register servers, run, interpret failures, export reports.

What you need

  • An MCP-capable assistant: Claude Code, Claude Desktop or Kiro.
  • Docker, to start the local test server and runner (optional if you already have a runner and a server).
  • Java 17 or Docker, to run the MCP server itself. Mac users without a JDK should use the container image.

The five-minute path

  1. Install the server and the skill (on a Mac, follow the Quick start on a Mac).
  2. Ask the assistant to check the machine: "Run frog_doctor and tell me what is missing."
  3. Ask for a local environment: "Start a local test server and runner." It will show you a plan and wait for your approval before starting containers.
  4. Ask it to run the example: "Scaffold a Test IG called hello in my workspace and run the starter test against the local server."
  5. Ask it to explain: "Why did that fail?"

Workflows walks through both personas end to end with the hello Test IG.

What you stay in control of

The server cannot write outside your workspace, never takes a secret as a plain tool argument, and asks before anything that starts a container, overwrites a file or writes to a shared server. See the safety model.

Release status

v0.1 covers environment, authoring, scaffolding, local and remote runs, server registry and report summaries. Publishing a Test IG, generating tests from an IG, rendering tests into IG pages and native installers are planned for v0.2 and v0.3. Planned tools are marked as such in the tool reference; nothing in the docs implies they exist today.

Pages