Document assembly, corpus retrieval,
and diligence — ready for attorney sign-off.

01Overview

This is a production-grade document assembly, corpus retrieval, and diligence engine designed to sit inside a private-capital operator's workflow. It runs as a local MCP (Model Context Protocol) server on port 9077, exposing 25 tools that generate real legal artifacts, run parallel background checks against public federal databases, and package everything with an attorney-review manifest.

It is not a lawyer replacement. It is a lawyer enabler — the work that used to take many hours of associate time on drafting and diligence for a typical SPV formation and Reg D 506(c) raise, this system does in seconds. The attorney then reviews the packet, signs the manifest, and files.

02The Boundary — UPL Discipline

The line between what this system does and what a licensed attorney does is the load-bearing design decision of the entire project. Every US state prohibits the unauthorized practice of law (UPL) — and every attorney-general's office in the country is actively enforcing it against automated tools that claim to provide legal advice. This system is built to stay on the right side of that line, every time.

03System Architecture

Six functional layers stacked on top of the FTH MCP Hub, each independently testable.

04MCP Tools

25 tools grouped by function. All accessible via POST /mcp/invoke.

Corpus retrieval  6 tools

Diligence  4 tools

Templates & assembly  6 tools

Filings & compliance  6 tools

Review & workflow  3 tools

05Template Library

17 templates covering WY LLC formation, Reg D 506(c) subscription documents, and a full 12-section Private Placement Memorandum. Every template carries a license badge and cites its regulatory basis in front matter.

Wyoming LLC formation packet

Reg D 506(c) subscription documents

Private Placement Memorandum (12 sections)

All PPM sections composable via legal_document_compose. All license SEC Filing Public — structural patterns drawn from actual SEC EDGAR-filed private placements.

06SPV-in-a-Box — Flagship Workflow

A single tool call produces the full formation + offering packet for a WY-formed Reg D 506(c) SPV. Given the deal facts, it composes documents, runs parallel diligence, generates the Blue Sky plan, registers the entity for compliance tracking, and writes review manifests for attorney sign-off — all in seconds.

1. Diligence in parallel with assembly

   The system kicks off diligence on Manager and Issuer concurrently with document composition. Total wall time is bound by the slowest DD provider, not sequential steps.

2. Formation packet

   Assembles Articles of Organization, Manager-Managed Operating Agreement, and Initial Written Consent. Each fact insertion is provenance-tagged for the manifest.

3. Subscription documents

   Assembles the Reg D 506(c) Subscription Agreement and Rule 501(a) Accredited Investor Questionnaire, correctly citing the verification framework of 506(c)(2)(ii).

4. Private Placement Memorandum

   Composes the 12-section PPM in a single pass. Asset-class-specific risk blocks are conditionally included based on the assetClass, usesLeverage, usesDigitalAssets, and hasInternationalComponent flags.

5. Blue Sky manifest

   Generates the state-by-state notice-filing plan for every state where investors reside. Includes fees, deadlines, EFD availability, and state-specific quirks (e.g. NY Martin Act broker-dealer position, FL cover-page legend).

6. Entity registration & compliance calendar

   Registers the newly formed entity and computes all upcoming ongoing obligations — WY annual report, Form D annual amendment, BOI, K-1 delivery targets, IRS 1065, annual OFAC re-screen.

7. Review manifest for every document

   Writes a Review Manifest (JSON + Markdown) for each document, listing template provenance, fact sources, cited regulations, attached diligence packets, and the attorney sign-off block. This is the artifact the attorney signs.

07Diligence Layer

Seven parallel providers, one unified severity model, red-flag surfacing.

Unified severity model

Every provider returns a DiligenceHit with severity. The orchestrator computes overall packet severity as the max, flagging incomplete if half or more providers errored.

08Regulatory Corpus

24 regulations pre-indexed with pinpoint citations. Direct fetch via eCFR API returns current text with retrieval timestamp for provenance.

09Blue Sky Coverage

9-state reference dataset covering ~80% of the deals a typical US operator sees. Every state entry is a starting point — the system flags that fees and deadlines must be confirmed against each administrator's current published schedule before filing.

10Compliance Calendar

11 obligation types tracked automatically per entity. Each obligation carries its regulatory basis, computed deadline, action steps, and estimated cost.

11Review Manifest — The Attorney Handoff

Every document generated by this system ships with a Review Manifest. This is the artifact the attorney reviews and signs. It contains everything an attorney needs to decide whether to rely on the document, and everything a court or regulator would want to see if the reliance is later questioned.

12Data Sources

Every regulatory and diligence data source is a public federal API. No vendor lock-in, no black-box aggregators, no legally-questionable scrapers.

Everything caches to disk with sane TTLs. First fetch may take a few seconds; every subsequent call to the same resource is instant.

13Quick Start

1. Install and start the hub

cd fth-mcp-hub
pnpm install
pnpm dev
# Hub listens on http://localhost:9077

2. Run a single tool

curl -X POST http://localhost:9077/mcp/invoke \
  -H 'Content-Type: application/json' \
  -d '{
    "tool": "legal_cfr_pinpoint",
    "arguments": { "citation": "17 CFR § 230.506(c)" }
  }'

3. Run the flagship SPV-in-a-Box workflow

curl -X POST http://localhost:9077/mcp/invoke \
  -H 'Content-Type: application/json' \
  -d @spv-input.json
# Returns full packet; documents saved to data/legal-ops/generated/spv-<id>/

4. Check compliance calendar

curl -X POST http://localhost:9077/mcp/invoke \
  -H 'Content-Type: application/json' \
  -d '{ "tool": "legal_compliance_calendar", "arguments": {} }'
# Returns upcoming obligations sorted by days-until-due

SEC_USER_AGENT=yourapp/1.0 ([email protected]) YourCo
COURTLISTENER_TOKEN=<from courtlistener.com/help/api/rest/>
SAM_API_KEY=<free from sam.gov/data-services/>

14Extending the System

15Repo Layout

fth-mcp-hub/
├── src/servers/
│   ├── legal-ops.ts               ← MCP dispatcher + tool manifest (25 tools)
│   └── legal-ops/
│       ├── types.ts               ← All shared types
│       ├── util.ts                ← HTTP + cache + hash + paths
│       ├── corpus-edgar.ts        ← SEC EDGAR client
│       ├── corpus-cfr.ts          ← eCFR / regulation index
│       ├── dd-ofac.ts             ← OFAC / Consolidated Screening List
│       ├── dd-gleif.ts            ← GLEIF LEI lookup
│       ├── dd-courtlistener.ts    ← Federal court records
│       ├── dd-sam.ts              ← SAM.gov entity + exclusions
│       ├── dd-finra.ts            ← FINRA BrokerCheck + SEC IAPD
│       ├── dd.ts                  ← Unified DD orchestrator
│       ├── templates.ts           ← Registry + assembly engine
│       ├── blue-sky.ts            ← State notice-filing manifest
│       ├── compliance-calendar.ts ← Ongoing obligation tracker
│       ├── manifest.ts            ← Review manifest generator
│       ├── spv-in-a-box.ts        ← Flagship workflow
│       ├── __smoke__.ts           ← Smoke tests
│       └── README.md              ← Operator reference
├── templates/legal/
│   ├── wy-llc/                    ← 3 templates
│   ├── reg-d-506c/                ← 2 templates
│   └── ppm/                       ← 12 templates
└── data/legal-ops/                ← Corpus cache · DD packets ·
                                    Generated docs · Manifests