# Clearon AI Legal Watch MCP

Review-gated, local-first monitoring for AI legislation, regulations, litigation, sanctions, court rules, and ethics guidance. It implements the two planned stages without publishing anything.

## Implemented scope

### Phase 1

- Unified candidate and authoritative-record schemas across six legal lanes
- Configured collectors for Federal Register, Regulations.gov, Congress.gov, and CourtListener/RECAP
- Explicit-source HTML scanning with SSRF protection and bounded responses
- Identity deduplication and mandatory human acceptance or rejection
- Material-field change detection, due-review queues, consolidated exceptions, and digests
- Append-only event log with a rebuildable snapshot and recoverable shared lock
- One-business-day high-priority and three-business-day standard candidate review SLAs
- Durable run metrics for source success, duplicates, review age, SLA breaches, false positives, and first discoveries

### Phase 2

- All state legislative discovery routes inherited from `trackers/state-legislation/jurisdictions.json`
- EU Cellar SPARQL and European Commission AI Office discovery routes
- Court-rules and sanctions discovery coverage that explicitly includes NCSC state-court materials, the Ropes & Gray AI Court Order Tracker, Damien Charlotin's AI Hallucination Cases Database, and U.S. Courts rules-committee agenda books
- Secondary legal-news discovery coverage for AI-and-law developments through LawSites, Artificial Lawyer, National Law Review, AI Lawsuit Tracker, Above the Law, Today's General Counsel, MultiState, StateScape, CAID, Orrick, Stanford sources, AI Law Tracker, and AI Policy Bulletin
- Capability-level verification: discovery never implies verified status or deep-document access
- Versioned official-document text with SHA-256 hashes and bounded comparisons
- JSON, CSV, and Markdown tracker exports
- Publication-neutral article-lead output, relationship fields, industries, topics, deadlines, and publication metadata

## Safety boundaries

- No WordPress or social mutation
- Search results are leads, not legal verification
- Only explicit review can move a candidate into the authoritative tracker
- Official instruments and dockets control over secondary trackers
- Retrieval failures become exceptions, never legal-status changes
- Active records must always have a next-review date

## Environment

Set `CLEARON_AI_LEGAL_WATCH_WORKSPACE` when the process is not started from the intended workspace root. Override the monitor root or legislation registry path explicitly with `CLEARON_AI_LEGAL_WATCH_ROOT` and `CLEARON_AI_LEGAL_WATCH_STATE_LEGISLATION_PATH` when needed.

`REGULATIONS_GOV_API_KEY` is required for Regulations.gov, `CONGRESS_GOV_API_KEY` for Congress.gov, and `COURTLISTENER_API_TOKEN` for CourtListener/RECAP. They may be provided through the environment or the protected local secret schema documented in `CREDENTIALS.md`. Federal Register and HTML collectors do not require keys. Missing keys make those collectors unavailable but do not affect the rest of the service.

## CourtListener access decision

Clearon uses the CourtListener REST API for scheduled legal-watch collection and keeps that path as the production default. The CourtListener MCP server is deferred for now: it may be useful later for ad hoc interactive docket, citation, or research sessions, but it is not part of the cron-critical monitoring path because isolated jobs are expected to use deterministic local commands instead of depending on inherited MCP tool surfaces.

## Typical workflow

1. `list_sources` and select a source with verified discovery capability.
2. `discover_source`, or `scan_explicit_source` for a known official URL.
3. Verify each lead against the controlling official record and call `record_candidate`.
4. Call `review_candidate` with an explicit reviewer and note.
5. Use `record_observation` and `record_document_version` for later changes.
6. Review `build_digest`, `get_due_reviews`, `get_operational_metrics`, and exceptions on a schedule. Follow `REVIEW_WORKFLOW.md`.
7. Use `export_tracker` or `list_article_leads`; publishing remains separate.

For the model-routed article workflow, see `EDITORIAL_HANDOFF.md`. Luna monitors, Sol verifies, Terra drafts, and Sol performs the final legal review; human approval and the existing draft-only editorial pipeline remain mandatory.

## Commands

```sh
cd tools/clearon-ai-legal-watch-mcp
npm test
npm run smoke
npm run collect -- --since=2026-08-03
npm run digest -- --hours=30
node --check server.mjs
node --check lib/watch.mjs
```

The `collect` and `digest` commands are the scheduler entry points. They call the same monitor library directly because isolated OpenClaw cron sessions may not inherit the host-visible MCP tool surface. Collection records structured, dated official-feed leads only; static HTML links remain discovery evidence and are not automatically queued.

For court rules, orders, sanctions, and FRE 707 monitoring, the configured discovery set now explicitly includes:

- `ncsc-ai-courts` for state-court AI rules, orders, and policies
- `ropes-gray-ai-orders` for federal and state court-order lead generation
- `damien-charlotin-hallucinations` for sanctions and hallucination-case lead generation
- `uscourts-rules-agenda` for Advisory Committee agenda books and federal evidence-rule developments such as FRE 707

For broader AI-and-law news discovery, the configured secondary-news layer now also includes:

- `lawsites-ai-legal-tech` for legal-tech and legal-AI reporting
- `artificial-lawyer-ai` for global law-firm and vendor AI developments
- `national-law-review-ai-law` for law-firm analysis and professional-responsibility developments
- `ai-lawsuit-tracker` for active AI-litigation lead generation
- `above-the-law-ai` for fast-moving BigLaw and legal-tech coverage
- `todays-general-counsel-ai` for the in-house legal and governance lens
- `multistate-ai` for state-legislation trend and topic discovery
- `statescape-ai-tracker` for StateScape AI-legislation discovery reference and authenticated report handoff checks
- `caid-state-ai-policy-tracker` for broader academic state-policy tracking
- `orrick-ai-law-center` for law-firm tracking of AI law and governance developments
- `stanford-codex-ai` and `stanford-hai-news` for academic and research-driven AI law/policy developments
- `ai-law-tracker-weekly` for broad free AI-law aggregation and weekly discovery leads
- `ai-policy-bulletin` for deeper AI policy and governance developments

Not every suggested free source is configured automatically. Some exact third-party tracker URLs are stale, blocked, or too JS-heavy from this environment, and official primary sources still control over all secondary discovery leads.

`NCSL` remains an approved reference source for legislation work, but its current site route is returning `403` from this environment, so it is not yet configured as an automated collector source.

For legislation-oriented secondary sources, the current collection layer has source-specific limits:

- `statescape-ai-tracker` now has a source-specific adapter, but the public route resolves to a marketing page plus a `printquicklink` handoff into an authenticated app report. In this environment the underlying report redirects to login, so the adapter returns explicit reference metadata and a warning rather than automated bill leads.
- `caid-state-ai-policy-tracker` now uses a dedicated structured-data adapter against the tracker’s published `data/*.json` assets and can emit legislation leads with bill metadata and official-source URLs. Those leads still require official-source verification before acceptance.
- `multistate-ai` is reachable, but the current public entry points are low-signal for automated bill discovery and may be better treated as human-reference or search-targeted inputs unless a better stable tracker route is identified.

So if these sources continue to return zero leads, prefer one of:

- a dedicated source-specific adapter,
- a targeted legislation-only sweep command,
- or keeping the source as manual/reference-only rather than forcing it through the generic collector.

## OpenClaw registration

```sh
openclaw mcp add clearon_ai_legal_watch \
  --command node \
  --arg /absolute/path/to/clearon-ai-legal-watch-mcp/server.mjs \
  --cwd /absolute/path/to/workspace
```

Scheduling is intentionally separate. Register and probe the MCP first, then create daily discovery and review jobs after their prompts and delivery destination have been approved.