Connect an AI assistant

The framework is available as a free, read-only MCP server. Connect it to an assistant that supports the Model Context Protocol (MCP) and it can look up controls, criteria and cross-references directly, instead of relying on what it remembers.

Server address

https://aacf.adriansaxton.com/mcp

No account, sign-in or key is needed. The server uses the Streamable HTTP transport.

Set it up

Claude (web and desktop)

Go to Customize > Connectors.
Click + Add, then Add custom connector.
Name it AACF and paste the server address above, then click Continue.
Under Authentication, choose No sign in, then click Add.

To use it in a conversation, click + at the lower left of the chat, then Connectors. Custom connectors are available on every Claude plan; the Free plan allows one.

Claude Code

claude mcp add --transport http aacf https://aacf.adriansaxton.com/mcp

VS Code

Run MCP: Add Server from the Command Palette, or add this to .mcp.json at the root of the workspace:

{
  "mcpServers": {
    "aacf": {
      "type": "http",
      "url": "https://aacf.adriansaxton.com/mcp"
    }
  }
}

Cursor

Add this to .cursor/mcp.json in a project, or to ~/.cursor/mcp.json to use it everywhere:

{
  "mcpServers": {
    "aacf": {
      "url": "https://aacf.adriansaxton.com/mcp"
    }
  }
}

Other assistants

Any client that supports remote MCP servers over Streamable HTTP can connect: give it the server address and no authentication.

Try asking

Which AACF controls cover ISO/IEC 42001 clause 6.1.2?
What does GOV-01 require at Level 3?
Which controls apply to agentic AI?
Run an AACF self-assessment of our governance with me.

What it provides

`get_framework_info`	Version, scale, overlays, owner roles, lifecycle stages, the standards cross-referenced and how to cite the framework.
`list_domains`	Every domain with the ID and title of each control in it.
`get_control`	One control in full: objective, levels 1 to 5, Level 3 criteria, questions, evidence, owners and cross-references.
`search_controls`	Find controls by words, domain, overlay, owner role or lifecycle stage.
`get_mapping`	Which controls cover an item of a standard, for example ISO/IEC 42001 clause 6.1.2 or EU AI Act Article 14.
`get_standard_coverage`	Every cross-referenced item of one standard and the controls covering each.
`get_assessment_questions`	Questions, evidence and the applicable Level 3 criteria for one domain, for a self-assessment.
`get_scoring_rules`	The assessor rulebook: how to place a level, evidence and how to report.

It also offers a self_assessment prompt that walks your assistant through an assessment, one domain at a time.

Privacy and limits

Read-only. The server only reads the published framework data. It stores nothing and does not log what you ask.
Self-assessment answers stay with you. The server supplies the questions and the scoring rules; your assistant does the scoring in your conversation, and your answers are never sent to the server.
Where the calls come from. With a Claude connector, Claude calls the server from Anthropic's cloud, not from your device. Other assistants may call it from your machine.
Which version. It serves a published release of the framework; get_framework_info says which. Every result carries the citation and licence (CC BY-SA 4.0).
Not a compliance opinion. Cross-references show where a control supports a source item. They are not certification or legal advice.
Free to run, so best effort. If the server is ever unavailable, the same data is on the downloads page.

The server's code is open source (MIT) on GitHub.