# MCP servers for GDPR compliance: what they do and what to check

URL: https://privacyboard.io/blog/mcp-server-gdpr-compliance

Published 2026-11-22, updated 2026-11-22.

An MCP server for GDPR compliance lets an AI assistant such as Claude, ChatGPT or Cursor read and work on privacy information through the Model Context Protocol. Some serve public reference facts, some reach your own compliance records. Before connecting one, check how it signs in, what it can change, where it runs and what it keeps.

An MCP server for GDPR compliance gives an AI assistant a direct line to privacy information, so it can answer from facts instead of memory. Some servers only serve public reference data, such as what a vendor's data processing agreement says. Others reach your own records: your vendor list, your requests, your register. The second kind is the useful one, and the one to check carefully before you connect it.

## What is an MCP server, in plain terms?

The Model Context Protocol (MCP) is an open standard that lets an AI assistant call tools hosted by someone else. An MCP server publishes a list of tools, such as "list my vendors" or "look up this company's DPA", and the assistant decides when to call them during a conversation. Claude, ChatGPT, Cursor, VS Code and many other clients speak it.

There are two ways a server runs:

- **Local**: a small program you install on your computer, which the assistant starts itself. It sees what is on your machine.
- **Remote**: a web address, such as `https://example.com/mcp`, operated by the provider. Your assistant connects to it over the internet, often after you sign in.

Since September 2025, servers can be published in the official MCP Registry, so assistants and directories can find them under a verified name.

## What can a GDPR MCP server do for a small team?

The useful jobs are the ones where the assistant otherwise guesses:

| Job | Without a server | With a server |
|---|---|---|
| "Does this SDK need a DPA? Where is its data stored?" | The assistant answers from training data, which may be out of date | It reads the vendor's current facts, with the source page |
| "Which privacy requests are due this week?" | You open the tool and look | It reads your request log and lists the deadlines |
| "Fill in this customer's security questionnaire" | It drafts generic answers | It drafts from your actual records and flags what is missing |
| "Add a cookie banner and privacy policy to this site" | It writes a generic template | It sets them up from the tools the code really uses |
| "Is anything missing from our register?" | It cannot know | It compares your activities with your vendors |

The value is not that the assistant knows the GDPR better. It is that it works from your records, so its answers can be checked.

## What kinds of GDPR MCP servers exist today?

As of 22 November 2026, the servers we found fall into three groups. This is not a ranking; each does a different job.

- **Guidance servers you run locally.** For example, `gdpr-shift-left-mcp` on PyPI (MIT licence, version 0.4.0 of February 2026) offers 34 tools for article lookups, DPIA and register workflows and code scans, aimed at developers using VS Code. `gdpr-compliance-ai-mcp` (CSOAI, version 1.1.16 of September 2026) offers tools such as a lawful basis assessment and a DPIA generator, with optional pay-per-call billing. Both describe themselves as guidance, not legal advice.
- **Public reference servers.** A remote server anyone can call without an account, serving facts such as vendors' agreements, roles and data locations.
- **Account connectors.** A remote server that signs you in and reaches your own compliance workspace: your vendors, requests, register, incidents and documents.

Privacyboard runs one of each of the last two: `https://privacyboard.io/mcp/vendors` (no sign-in, facts on 1250+ SaaS tools and a cookie banner setup from a project's packages) and `https://privacyboard.io/mcp` (your workspace, on every plan). Both are in the official registry under `io.privacyboard`.

## What should you check before connecting one?

An account connector is a new way into personal data, so treat it like any other vendor. The GDPR questions are the usual ones, applied to a new kind of tool.

**How does it sign in?** A remote server that reaches your records should use OAuth, the sign-in flow the MCP specification describes, so you approve access in your own account and can revoke it. Be wary of a server that asks you to paste a long-lived key into a configuration file. Security research reported by the IAPP in September 2026 found many exposed servers accepting requests with no authentication at all.

**What can it change, and who confirms?** Reading is one thing; changing records is another. Prefer a server that proposes changes and applies them only after you confirm each one, and that lets you grant read-only access.

**What does the assistant receive?** Everything a tool returns goes into the conversation, and so to the assistant's provider. Apply data minimisation (Article 5(1)(c) GDPR): a good connector returns what the question needs and leaves out what it does not, such as the names and messages of the people behind a privacy request.

**Who is the processor?** If the server holds or processes personal data for you, its operator is a processor and you need a data processing agreement (Article 28). The AI assistant's provider is a second vendor in the chain, with its own terms.

**Where does it run, and what does it keep?** Ask where the server is hosted, whether it logs requests and for how long, and whether data leaves the EU. Logs of tool calls can themselves contain personal data.

**Is it really from the company it names?** Use the address from the provider's own documentation or from the official registry, where the name is tied to a verified domain.

## Checklist before you connect a GDPR MCP server

- [ ] The address comes from the provider's documentation or the official MCP Registry
- [ ] It signs in with OAuth and can be disconnected from your account
- [ ] Read-only access is possible; changes need your confirmation
- [ ] You know which data its tools return to the assistant
- [ ] A DPA covers it if it processes personal data for you
- [ ] You know where it is hosted and how long it keeps logs
- [ ] It is listed in your vendor list and, if it processes personal data, in your register

## What to do this week

1. List the assistants your team already uses for work, and whether any is connected to tools that hold personal data.
2. Connect one read-only compliance source to the assistant you use most, and ask it the three questions you answer most often.
3. Add the connector and the assistant's provider to your vendor list, with their agreements.
4. Write down who on the team may connect new servers, so it does not happen silently.

## FAQ

### Is connecting an MCP server to Claude or ChatGPT GDPR compliant?
It can be, like any other tool: it depends on what data flows, under which agreements, and whether you could explain it to a customer. The checks above are the same ones you run for any new vendor.

### Do I need a DPA with the MCP server's provider?
If the server stores or processes personal data on your behalf, yes: its operator acts as your processor. A server that only serves public reference facts, without receiving personal data, usually does not need one.

### Can an AI assistant change my compliance records through MCP?
Only if the server offers tools that write, and you allow them. Look for servers that show each change and wait for your confirmation before applying it.

### What is the difference between a local and a remote MCP server?
A local server runs on your computer and sees your files; a remote server is a web address run by the provider, usually behind a sign-in. For records you share with a team, a remote server with proper sign-in is easier to control and to revoke.

### Where can I find MCP servers for privacy work?
The official MCP Registry lists servers under verified names, and directories such as Glama rate their tool definitions. Search for the provider you already use before trying an unknown one.

## Sources

- [Model Context Protocol: introduction](https://modelcontextprotocol.io/introduction)
- [Model Context Protocol: authorization](https://modelcontextprotocol.io/specification/2025-06-18/basic/authorization)
- [The official MCP Registry](https://blog.modelcontextprotocol.io/posts/2025-09-08-mcp-registry-preview/)
- [GDPR, Articles 5 and 28 (EUR-Lex)](https://eur-lex.europa.eu/eli/reg/2016/679/oj)
- [IAPP: Building privacy-safe MCP servers, what GDPR actually requires (9 September 2026)](https://iapp.org/resources/article/building-privacy-safe-mcp-servers-what-gdpr-actually-requires)
- [gdpr-shift-left-mcp on PyPI](https://pypi.org/project/gdpr-shift-left-mcp)
- [gdpr-compliance-ai-mcp on PyPI](https://pypi.org/project/gdpr_compliance_ai_mcp/)
