MCP server · Developer tools
Groundwork MCP server
by nickjlamb
Lets your AI run readiness checks on document QA and extraction projects, and scaffold the checks into a repo.
Groundwork is a toolkit for people building AI systems that read documents and answer questions or pull out fields. It helps you check that your system is not making things up, that private details are hidden first, and that a small change does not quietly break everything. It is handy if you have a prototype that reads documents and you want to know whether it is safe to put in front of real people.
What is an MCP server? The 30-second version
On its own, your AI can only chat with you. An MCP server is a small helper program that gives your AI a new skill. This one, Groundwork, lets your AI run readiness checks on a document QA or extraction project, and even set up the checks inside a project folder. So instead of you running commands yourself, you can just ask your AI and it does the checking for you.
What this MCP server does
You ask your AI to check a project or a single answer. The AI passes that request to the Groundwork helper. The helper runs deterministic checks (the same input always gives the same result): it tests redaction, checks whether numbers in an answer actually appear in the source, checks whether the system abstains when it should, and for extraction it checks fields against a schema and against hand-labelled gold cases. You get back a pass or fail with the exact problem named, so you know what to fix.
What you can do with it
- Run a full readiness check on a project folder
- Check whether a single answer is grounded in the provided text
- Check a single extracted record for schema validity and fabricated fields
- Scaffold the readiness harness into a new project
- See measured token usage and cost savings for a project
- Freeze a good run as a baseline so future runs must not fall below it
Try asking your AI
- “Run a readiness check on my document QA project and tell me what failed”
- “Is this answer grounded in the source? Here is the answer and the text it came from”
- “Check this extracted record against my schema and flag any guessed fields”
- “Scaffold a Groundwork harness for a new extraction project in this folder”
- “Show me the cost summary for this project”
What it gives back to you
You get back a plain result in the chat: either all checks passed, or a list of failures with names like ungrounded number "14" or FABRICATED field "date_of_birth". For cost questions you get token counts and where the savings came from. Nothing is changed in your system unless you ask it to scaffold a harness, which writes files into your project folder.
Before you start
What you need
- Node.js installed (18 or newer)
- The Claude desktop app or Claude Code
- A project folder for the checks that look at a repo (single-answer and single-record checks work without one)
Good to know
Scaffolding writes new files into your project folder, and a readiness check can freeze a baseline that later runs must not fall below, so review the changes before committing them.
Install it with your AI
Add Groundwork MCP server to your AI, no technical skills needed
You don't install anything by hand. You copy one prompt, paste it into an AI that can work on your computer, and it checks, installs and connects the server for you, asking you when it needs something.
Sign in to get the install prompt
Members get a ready-made prompt that lets the Claude desktop app check Groundwork MCP server, install it and connect it for them, step by step. You don't need any technical skills: you copy, paste and answer a few questions. Your connected AI can also find and install any of the 4,361 MCP servers here for you.
Who it's for
People building or maintaining AI systems that read documents, such as analysts, ops teams, and product folks working with engineers on document QA or data extraction.
