ComplyLayer

AI governance and EU AI Act compliance from your AI assistant. Check your AI inventory, risk classification and missing documents, assess a planned AI use, and propose AI tools found in Slack, GitHub or Notion for review.

Hosted MCP Server

npx add-mcp 'https://app.complylayer.com/api/mcp/'

Installs into Claude Code, Codex, Cursor and more

Documentation

Your AI assistant finds your AI and tells you where you stand

Connect Claude, Claude Code or Cursor to ComplyLayer. Your assistant searches the tools it is connected to, like Slack, Notion, Drive, Jira and GitHub, with its own access, and proposes every AI tool and use case it finds. You review each one before it enters your inventory.

EU AI Act · Art. 26GDPR · Art. 30

Free plan, no time limit · No credit card · Nothing filed without your approval

claude.ai

Claude ComplyLayer connected

Find the AI our team uses and propose it to ComplyLayer.

Searching the tools I am connected to

ChatGPT

Seen in #marketing in Slack

Proposed

GitHub Copilot

Seen in 3 repositories

Proposed

Otter.ai

Seen in Sales calls in Notion

Proposed

HireScore

Seen in #hiring in Slack, screens job applicants

Proposed

Sent 4 tools to ComplyLayer. Nothing is added until you accept it.

The problem

The inventory is only as good as what people remember to add

AI use spreads through Slack threads, repositories and side projects faster than anyone fills in a form: an agent wired to the data warehouse, an LLM call in a support tool, a coding assistant with database access. Your AI assistant can already see those places. Now it can report what it finds, and developers can check a new AI use before they build it.

✓AI inventory3 tools, updated in March

MP

#sales

Fireflies writes up all my client calls now

not in inventoryJS

#eng

pushed the Gemini summariser to prod

not in inventoryLT

#hr

trying an AI CV screener this week

not in inventory?

You, to yourself

Who else is using AI that nobody told me about?

The solution

How ComplyLayer solves it

  1. 1AI tools and use cases found where they actually appear, with a link to the source
  2. 2What each one connects to, with what access, and whether it acts on its own
  3. 3A review queue, so nothing enters the inventory without a person
  4. 4check_usage answers "may I send this data to that model?" from your own setup

claude.ai

Claude ComplyLayer connected

Find the AI our team uses and propose it to ComplyLayer.

Searching the tools I am connected to

What's included

Connect it

  • Works with Claude, Claude Code, Cursor and any MCP client
  • Sign in with OAuth, or use a personal token
  • One click from the Claude connector directory, on every plan
  • Disconnect any assistant in one click

It finds AI

  • Built-in prompt to scan your workspace for AI
  • Finds tools, use cases and connected systems, with where each was seen
  • Proposals capture connected systems, data, autonomy and evidence
  • Accepting reclassifies the tool and never lowers a risk signal
  • ComplyLayer never reads your Slack, GitHub or docs

Ask where you stand

  • Ten tools: inventory, compliance overview, gaps, alerts, documents, acknowledgements, usage check and two kinds of proposal
  • Ask where you stand and get your score and gaps, with links to the screen that fixes each one
  • Quotes your score, gaps and alerts. Nothing changes until you accept it

The outcome

Every proposal waits for a person

0

sources ComplyLayer reads

10

tools for your assistant

1 click

to accept or dismiss

Accepted

app.complylayer.com/systems

Proposed by your assistant3 waiting

Fireflies.ai

Seen in Slack #sales · 3 messages

LimitedAcceptDismiss

Gemini API

Seen in GitHub acme/summariser

LimitedAcceptDismiss

HireScore

Seen in Notion · HR hiring plan

HighAcceptDismiss

  1. 1

    Found where it is used, with the source

    Each proposal links to the Slack thread, repository or document where the tool appeared, so whoever reviews it can check before accepting.
  2. 2

    Nothing enters without a person

    Your assistant proposes and a person in ComplyLayer accepts or dismisses. Accepting never lowers a risk signal.
  3. 3

    Answers from your own workspace

    Ask where you stand and the figures come from your register, documents and alerts, with a link to the screen that fixes each gap.

Frequently asked questions

What is the ComplyLayer MCP server?

A Model Context Protocol server that lets your own AI assistant work with ComplyLayer. It can read your AI inventory and where you stand on compliance, check a planned AI use against your framework, and propose AI tools and use cases it finds in the tools it is connected to, like Slack, Notion, Drive, Jira and GitHub, for you to review.

Does ComplyLayer read our Slack, GitHub or Notion?

No. Your assistant searches those sources with the access it already has through its own connectors. ComplyLayer only receives a proposal: the tool, what it is used for, what it connects to, and a link and one-line summary of where it was seen. No message text, secrets or personal data.

Which assistants work with it?

Claude on claude.ai and the desktop app (ComplyLayer is listed in the Claude connector directory), Claude Code, Cursor, and any MCP client that supports Streamable HTTP. Claude and Claude Code connect by signing in to ComplyLayer; other clients can use a personal token.

Can the assistant change our inventory?

No. Everything it finds lands in a review queue. A person accepts or dismisses each proposal, and accepting only ever adds a use case or raises a risk signal. The assistant cannot approve documents, classifications or proposals.

How do developers use it?

Before building or wiring up AI, a developer asks their assistant, for example whether customer data may go to a model or an AI agent may write to the warehouse. The check_usage tool answers from your own setup: whether the tool is registered, the projected risk tier, the documents that use needs, and what to check, such as a DPA, a DPIA or human approval for write access.

How is access secured?

Sign-in uses OAuth 2.1 with PKCE. Access tokens last one hour and refresh tokens rotate, with reuse detection. A connection acts as the person who allowed it, only inside their organization, and can be disconnected at any time. Viewers cannot connect an assistant.

Which plans include it?

Every plan, including Free.