- Home
- AI development
- MCP servers
- AI use policy server
Public MCP server
Draft your company's AI use policy from inside your AI assistant
This is a public MCP server, free to use, with no account or sign-in. Connect it to Claude or another MCP client, answer eight short questions, and the assistant returns a short AI use policy draft in Markdown. We run it as a working example of the servers we build. IG Digital Lab is a technology provider, not a law firm: the draft is not legal advice, so have your counsel review it.
- Prices published on the site
- First working version in two weeks
- Support month to month, cancel any time
Eight questions in, a ten-section draft out
The server has two tools. get_policy_questions returns eight questions: company name, industry, the AI tools in use, the kinds of data the team handles, allowed uses, prohibited uses, who approves new tools, and how often the policy is reviewed. draft_ai_use_policy takes the answers and returns a short policy, about two pages, in Markdown.
The policy has ten short sections: purpose, scope, approved tools, data rules, allowed uses, prohibited uses, human review of AI output, approval of new tools, incident reporting and the review date. The data rules always name client confidential information, personal data and credentials as things that never go into a free or unapproved AI tool, and add whatever else you list.
The draft is a fixed template filled in with your answers. No AI model writes it, so the same answers give the same text. It is a starting point for internal use. Your counsel should review it before you adopt it, and a regulated industry will need more than a template can say.
Two tools, called in this order
get_policy_questions
Returns eight questions with limits and an example for each. It takes no input. The assistant calls it first so that it asks you the right things.
draft_ai_use_policy
Takes the answers and returns the policy as Markdown, with the effective date, the next review date and a list of the parts still to fill in. Only the company name is required. Every other answer left out becomes a bracketed placeholder. After the policy, the result lists them, and flags any free plan or kind of data that needs counsel's check.
Add the server to your assistant
The address is https://igdigi.com/mcp/ai-policy. It needs no account, key or sign-in.
claude mcp add --transport http ai-use-policy https://igdigi.com/mcp/ai-policy{
"mcpServers": {
"ai-use-policy": {
"type": "http",
"url": "https://igdigi.com/mcp/ai-policy"
}
}
}curl -s https://igdigi.com/mcp/ai-policy -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Claude on the web or desktop, on a Free, Pro, Max or Ultra plan: open Customize > Connectors, choose Add custom connector, paste the address and, if asked about authentication, choose No sign-in. A Free plan can have one custom connector. On Team and Enterprise plans an Owner adds it under Organization settings > Connectors > Add > Custom, with the type Web, and each member then connects from Customize > Connectors. Other MCP clients, including Cursor and VS Code, accept a remote server address in their MCP settings, and the field names differ by client. Claude steps checked against the Claude Code and custom connector documentation on 2026-11-07.
What a call returns
The assistant sends the answers as arguments. This example uses a made-up company, and it leaves out allowed uses and prohibited uses, which the draft then shows as bracketed placeholders.
{
"company_name": "Example Design Studio",
"industry": "graphic design",
"ai_tools": ["ChatGPT (business plan)", "Claude (team plan)"],
"data_handled": ["Financial records and payment details"],
"approver": "Office manager",
"review_cadence": "every six months",
"effective_date": "2026-11-08"
}# AI Use Policy: Example Design Studio
Effective November 8, 2026. Next review May 8, 2027. Policy owner: Office manager. Industry: graphic design.
## 1. Purpose
Example Design Studio uses AI tools in its work. This policy sets which tools are approved, what data stays out of them, who checks the output, how new tools get approved and what to do when something goes wrong.
## 2. Scope
Everyone who does work for Example Design Studio, on any device or account. "AI tool" means chat assistants, writing and coding assistants, AI features inside other software, and AI agents that take actions. Legal, contractual and professional duties that already apply to Example Design Studio still apply.
## 3. Approved tools
- ChatGPT (business plan)
- Claude (team plan)
Use them with a company account only. Anything not listed needs approval under section 8 first.
## 4. Data rules
Never enter into a free, personal or unapproved AI tool:
- Client or customer confidential information
- Personal data about customers, employees or applicants
- Passwords, API keys and other credentials
- Anything covered by a confidentiality agreement or a contract that limits sharing
- Financial records and payment details
In an approved tool, use this data only when the task needs it. Share the minimum: remove names, account numbers and other identifiers when the task still works without them.
Data that a law, a professional rule or a contract protects, such as patient health information, client confidences or tax return information, goes into an AI tool only after counsel confirms that the agreements and consents that rule requires are in place.The result continues with allowed uses, prohibited uses, human review of AI output, approval of new tools, incident reporting and the review date. The policy ends with two lines: "Draft for internal use, not legal advice. Review it with your counsel." and "Generated by IG Digital Lab: https://igdigi.com/ai/mcp-development/ai-use-policy-server". After them comes a list headed "Still to fill in or check": for this call it flags the payment data for counsel, plus every answer left out.
What happens to what you type
Kept only while the request runs
The server builds the draft while your request is open and sends it back. It writes your answers to no file, no database and no log line.
No AI model on our side
The server fills a fixed template with plain code and calls no AI model. Your answers do pass through the assistant you use, such as Claude, under that assistant's own terms. The server adds no other model provider and receives only the answers.
Ordinary web traffic
The request travels over HTTPS through Cloudflare, like every page of this site. The web server's access log records that a request was made, with the usual time, address and result code, and not what it contained.
Who should use it and who should skip it
A good fit when
- Your team already uses ChatGPT, Claude or Copilot and nobody has written down what may go in
- You have no AI policy yet and want a structured first draft in one sitting
- You want to see a working MCP server before you ask anyone to build one
Something simpler will do when
- Your industry has its own rules about client or personal information: your counsel writes that part, and a template will not cover it
- You need a policy tied to a contract, an audit or an insurance questionnaire: someone has to read those documents
- You already have a policy: use the questions as a checklist at its next review
This server is free. One on your own systems is quoted after the free audit
One workflow running end to end, or one internal tool.
Several systems made to work as one: integrations, roles, reports.
A server over your database, CRM or files is quoted after the free audit: one process ($2,500 to $6,000) for a read-only server over one system, connected systems ($6,000 to $15,000) with per-person sign-in and several sources. See MCP server development for what we build, and how we keep a server read-only by default.
Care $300 to $500 a month. Active $750 to $1,500 a month. AI model usage, GPU time and hardware are paid to the provider, on accounts in your name.
Questions about the AI use policy server
Does the server use an AI model to write the policy?
No. The policy is a fixed template filled in with your answers by plain code. The same answers on the same date give the same text. The server sends your answers to no AI model; the assistant you type them into handles them under its own terms.
Is anything I type stored?
Not by this server. It builds the draft in memory and sends it back, and it writes your answers to no file, database or log. The request passes through Cloudflare, like all traffic to this site, and the web server's access log records that a request was made, not its content.
Is the draft legal advice?
No. IG Digital Lab is a technology provider, not a law firm. The draft is for internal use, and it says so at the bottom. Have your counsel review it before you adopt it, especially if your industry has rules about client or personal information.
Which assistants can use it?
Claude, through Claude Code and custom connectors, and other tools that accept the address of a remote MCP server. Whether a product or plan allows custom servers is the maker's decision, so check its settings.
Can I call it without an AI assistant?
Yes. Any program that can send a JSON request can call it, and the curl example above lists the tools. An assistant is easier: it asks the questions and reads the result back.
Can you build a server like this for our company?
Yes. It starts with the free written audit, which we send by email within two business days, and the quote follows: one process ($2,500 to $6,000) for a read-only server over one system, connected systems ($6,000 to $15,000) with per-person sign-in and several sources.
Related
- MCP servers Connect AI assistants to your own systems
- All services Seven things we build and run
- AI agent setup Your own AI agents, installed and taught
- AI readiness assessment A written review of where AI fits
- AI for small business One task automated first, priced up front
- AI development Every AI service we build, on one page
Tell us what's slowing you down.
You get a free written review by email within two business days.