Model Context Protocol
The corpus, inside the assistant you already have open.
Nobody answers a questionnaire in a browser tab any more. This is the same guidance our question pages carry, served over MCP so your assistant can search it while you are in the spreadsheet. It is free, read-only, and needs no account.
Add it
One HTTP endpoint, nothing to install and nothing to keep updated. There is no package to fall behind the corpus.
Claude Code
claude mcp add --transport http filltrust https://www.filltrust.com/api/mcpAny client that takes a config file
{
"mcpServers": {
"filltrust": {
"type": "http",
"url": "https://www.filltrust.com/api/mcp"
}
}
}Four tools
What it can be asked
search_questions
Paste the row from the spreadsheet. Comes back with the questions in the corpus that match it, each with the address of the page it came from.
get_question
The whole of one page: what the question is really asking, which of your documents answers it, what evidence to attach, a model answer with the specifics left blank, and the ways it usually goes wrong.
list_standards
Which standards the corpus covers and which controls it cites, with the real title of each. SIG Lite and VSA Core are named without numbers, because neither publishes a control list anyone can check.
get_filltrust_posture
Our own answers, for an agent assessing us as a vendor. Including the ones that are no.
Worth knowing
What it will not do
It does not read your documents. It serves pages we have published. Drafting an answer from your own policies and citing the passage it came from is the product, and it needs an account.
It does not write an answer for you. Every model answer it returns is a template with the specifics deliberately blank, and it says so in the payload. An assistant that fills those in from nothing is asserting a control you may not have.
It does not invent a control number. References come from an allowlist carrying the verified title of each control, and the build fails on anything else. Roughly half of ours were wrong once, which is why that allowlist exists.
Then do the other two hundred rows properly.
The corpus tells you how a question should be answered. The product answers it from your documents and shows you the passage it used.