AI regulation MCP server for assistants
An open-source Model Context Protocol server that lets an AI assistant look up AI laws, obligations, deadlines and enforcement actions here, instead of scraping pages. Read-only, no key, and every result carries the record's official source and review status.
How do I install it?
It runs on your machine with Node.js and calls the public site; there is no hosted endpoint yet. Get the code, then register the server with your MCP client:
git clone https://github.com/8h45k4r/aipolicytracker.git
{
"mcpServers": {
"aipolicytracker": {
"command": "node",
"args": ["/path/to/aipolicytracker/agent/server.mjs"]
}
}
}
No dependencies to install. Source and tests: agent/ in the repository.
What can an assistant do with it?
| Tool | Returns |
|---|---|
| corpus_health | How far the corpus can be trusted right now: records published, overdue for re-check, verified by a named reviewer |
| search_policies | Instruments by keyword, jurisdiction or status |
| resolve_record_id | A permanent id (AIPT-POL-0001) to the record and its current address |
| get_policy | One instrument as a Markdown context file, provenance included |
| get_jurisdiction | One jurisdiction profile as a Markdown context file |
| list_obligations | Duties recorded against instruments, with the provision each comes from |
| recent_changes | The dated change log, with a significance score |
| search_incidents | AI incidents, with the laws on record that address the harm |
| build_obligations_register | The applicability check as an obligations register, with XLSX, CSV and PDF links |
| get_applicable_deadlines | Which dates apply to one organisation, and why |
| list_transition_measures | AI economic transition measures |
| get_displacement_index | The displacement policy index per jurisdiction |
| list_enforcement_actions | Enforcement actions with regulator, respondent, legal basis and outcome |
| list_implementation_measures | What an instrument still depends on, with due and actual dates |
| list_templates | The templates library |
| open_gaps | What the corpus is missing |
corpus_health comes first on purpose: most records are not yet confirmed by a named reviewer, and an assistant should say so before quoting one.
Prefer plain HTTP?
The same data is in the read-only API and the bulk exports. For higher limits and service terms, see plans.
Frequently asked questions
Is there an MCP server for AI laws?
Yes. AIPolicyTracker publishes an open-source MCP server with 16 read-only tools for AI policy records: search instruments, read a jurisdiction, list obligations and deadlines, enforcement actions and changes. It runs on your machine over stdio and calls the public site.
Is there a hosted MCP endpoint?
Not yet. Today the server runs locally with Node.js and needs no account or key. A hosted endpoint with higher limits is planned for the paid plans.
Does the MCP server show whether a record is verified?
Yes. Its corpus_health tool reports how many records were confirmed by a named reviewer, and every tool result carries each record's official source and review status, so an assistant does not present a summary as settled law.