Skip to content

About

BrokerFlow broker operations platform with authenticated ChatGPT MCP tools, a private plugin, review workflows, and funding-safe automation.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

199 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Image

BrokerFlow AI

This repository contains the BrokerFlow broker operations platform, its public site, and the existing Loan Broker Automation Architect GPT Actions. Public pages resolve at /, /agents, /actions, /knowledge, /workflows, /plugin, and /docs; deployment routes are covered by tests/site/public-routes.test.js. The responsive operations console lives at /app/* and is implemented in app.html, app.css, app.js, and data.js. The console uses clearly labeled sample records until an authenticated data source is configured. Existing API Actions, authenticated MCP tools, workflows, and knowledge files remain distinct integration surfaces.

Compliance note: BrokerFlow AI recommends, drafts, prepares, routes, and logs. It never approves, declines, qualifies, guarantees, underwrites, or funds. All site copy uses review-safe language ("potential fit", "ready for broker review", "possible lender match", "based on information provided") and never says "approved", "guaranteed", "qualified", or "eligible".


Site URL
Web https://brokerflow-ai.vercel.app/
ChatGPT https://chatgpt.com/g/g-6a214c8c41d48191bebafa39c853412f-loan-broker-automation-architect
Notion https://github.com/JFeimster/brokerflow-ai/
Vercel https://vercel.com/jason-feimsters-projects/brokerflow-ai
GitHub https://github.com/JFeimster/brokerflow-ai/

Action packs

BrokerFlow also contains no-auth GPT Action packs for intake, scoring, routing, workflow triggers, and controlled review queues.

Action Operation ID Vercel endpoint Schema
Lender Match Review requestLenderMatchReview POST /api/no-auth/lender-match-review schemas/no-auth-lender-match-review.schema.yaml
Automated Lender Fit Routing routeLenderFitScenario POST /api/no-auth/lender-fit-routing schemas/no-auth-automated-lender-fit-routing.schema.yaml

The 12 partner lifecycle Actions have a consolidated endpoint and operation map in docs/internal-partner-action-packs-overview.md, plus GPT Builder setup, Vercel environment configuration, and testing guidance.

Automation boundary

The legacy REST lender-fit routing Action retains its existing automation response contract. The new MCP route_lender_fit tool uses the shared deterministic calculation without triggering outbound effects. Internal scores remain heuristics, not underwriting or current lender offers.

Human review should only block final regulated steps: approval, denial, underwriting, formal pricing, funding, or lender submission. The lender-match review action exists for cases where a manual queue is required.

Folder structure

site/
  README.md            This file — site overview, design system, and authoring rules
  pages/
    home.md            Landing page: hero, value prop, how it works, demo, CTA
    agents.md          The GPT agent: role, guardrails, what it does and won't do
    actions.md         Connected actions: CRM sync, pipeline, documents, tasks
    knowledge.md       Knowledge base: how BrokerFlow reasons and what it knows
    workflows.md       Automations: n8n, Zapier, Make recipes and triggers

Release notes

See CHANGELOG.md for the private plugin v1.3.0 activation record and v1.3.1 source update. The current authenticated release state and follow-up checks are tracked in docs/MCP_RELEASE_CHECKLIST.md.

Audience

  • Independent loan brokers and small brokerages who want an AI operations layer.
  • Funding shops (e.g. Moonshine Capital / Distilled Funding) standardizing intake and CRM hygiene.
  • Partners evaluating whether BrokerFlow fits their lender-submission process.

Design system

The site uses a premium dark AI/SaaS aesthetic. Page copy here is layout-aware: each page maps to the standard section flow and references these tokens so the front-end stays consistent.

Color tokens

Token Value Use
bg #070A12 Page background
surface #101827 Cards, panels
surface2 #151F32 Nested panels, code, table rows
text #F8FAFC Primary text
muted #94A3B8 Secondary text, captions
primary #7C3AED Primary buttons, key accents
secondary #06B6D4 Links, secondary accents
hot #F97316 Highlights, badges
success #22C55E Positive states, checks
border rgba(148,163,184,0.18) Hairline borders

Standard landing flow

Every page composes from this section order (omit sections that do not apply):

  1. Hero — one-line promise + primary CTA
  2. Built For — who it is for
  3. What It Does — core capabilities
  4. How It Works — step sequence
  5. Demo / Embed — interactive preview or video
  6. Use Cases — concrete scenarios
  7. Pricing / Access — how to get it
  8. Proof — real results only (no fabricated metrics or logos)
  9. FAQ — common questions
  10. Footer — links, legal, compliance line

Authoring rules

  • Mobile-first: write short paragraphs and scannable lists; assume narrow viewports first.
  • Real proof only: never invent testimonials, client names, metrics, or partner logos.
  • Review-safe language everywhere: no approval/guarantee/qualification claims.
  • Front matter (optional) per page: title, description, slug, order for SEO + nav.
  • Keep one # H1 per page (the page title); use ## for sections and ### for sub-sections.
  • Link between pages with relative paths, e.g. [Actions](./actions.md).

SEO

  • Each page should declare a title (<= 60 chars) and description (<= 155 chars).
  • Use descriptive section headings; they become the page outline and anchor links.
  • Canonical domain: https://brokerflow-ai.vercel.app.

Build & deploy

  • Public pages are static HTML at the repository root and resolve through Vercel clean URLs. Editorial source remains under site/pages/.
  • Route and navigation checks live under tests/site/ and run with npm test.
  • Preview locally with npm run dev; Vercel Preview and production releases follow repository controls.

License

Proprietary / UNLICENSED. © Moonshine Capital / Distilled Funding. Internal and partner use only.

Authenticated MCP and private plugin

BrokerFlow includes a stateless Streamable HTTP MCP endpoint at https://brokerflow-ai.vercel.app/mcp, authenticated review and follow-up services, and a portable private plugin package under plugin/. Supabase OAuth 2.1 is active. The owner-uploaded plugin v1.3.0 connected in ChatGPT, where all seventeen tools were discovered and nine live read-only MCP calls succeeded. The current server and v1.3.1 source package add get_follow_up_task; the updated package still needs owner-managed private refresh and live verification. See docs/MCP_ARCHITECTURE.md, docs/MCP_OAUTH_SETUP.md, and docs/MCP_RELEASE_CHECKLIST.md. Build a validated Plugin Creator-ready ZIP with npm run package:plugin; packaging itself is not installation or live-authentication proof.

The server registers eighteen tools, including the seven core operations: capture_borrower_intake, route_lender_fit, request_lender_match_review, generate_document_checklist, trigger_missing_document_request, lookup_deal_status, and capture_partner_referral. New writes require explicit confirmation; routing/checklists are preliminary and document follow-up stops at a tracked draft/internal task. Canonical platform-core leads, referrals and audit events are reused. The additive core-workflows migration has been applied to the active platform-core project. See the inspection inventory, tool catalog, and synthetic examples.

npm test recursively discovers non-database tests under tests/, including MCP, REST compatibility, compliance and skill tests. npm run test:postgres runs both PostgreSQL suites with DATABASE_URL pointing to an isolated test server; the core suite creates and drops its own synthetic database. CI uses ephemeral PostgreSQL 17 and packages the private plugin. Never point integration tests at a shared or populated database.

The public pages and static-route release checks are described in docs/WEB_RELEASE_CHECKLIST.md. The plugin onboarding page is /plugin; private plugin deployment and shared OAuth/database activation remain owner controlled.

About

BrokerFlow broker operations platform with authenticated ChatGPT MCP tools, a private plugin, review workflows, and funding-safe automation.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Contributors

Languages