# MagicTeams Product Portfolio

> Every product we ship: 27 platforms, 26 MCP connectors, 3 AI revenue agents — from one GitHub org (everyai-com, 192 repos, 68 public).

https://magicteams-portfolio.pages.dev

Machine manifest: https://magicteams-portfolio.pages.dev/.well-known/agent.json

## Platforms (27)

### MagicTeams Voice + Text AI — AI calling, campaigns, calendars, client management

The comms + voice platform everything else builds on. AI phone receptionists answer 24/7, run outbound campaigns (one-off, bulk, browser demo), manage calendars and clients. Four voice engines per agent — Ultravox default with full in-call tool parity (transfer, booking, custom tools), Gemini Live, Sarvam (11 Indian languages), Cloudflare edge worker. Twilio/Telnyx telephony, transcripts with Gemini summaries, call outcomes, knowledge-base injection, minutes metering, plus a public API (mt_live_ keys) and an MCP server exposing voice, SMS, WhatsApp and email as agent tools.

Why it exists: Local businesses lose jobs every day to calls they miss after hours — and a human receptionist for every line is unaffordable. MagicTeams was built so AI answers every call, books the job, and runs follow-up campaigns without adding headcount.

Features: 24/7 AI phone receptionists on your numbers; 4 voice engines: Ultravox (full in-call tools), Gemini Live, Sarvam (11 Indian languages), Cloudflare edge; Outbound: one-off, bulk campaigns, browser demo calls; Calendars, client management, knowledge-base injection; Transcripts + Gemini summaries, call outcomes, minutes metering; Public API (mt_live_) + MCP server: voice, SMS, WhatsApp, email as agent tools.

How it helps: Never miss a call or lead again — a solo shop answers, books and follows up like a team of ten, and agents can drive the whole platform through the API.

Live: https://app.magicteams.ai

### MagicBots — Open-source team-of-bots chat app

Your own team of AI bots in a chat app — each bot a real agent (Claude, Codex or Pi) with its own model, computer and connected apps. Build a team where each bot has a role and together they get the work done. The voice workspace adds AI voice agents, scheduled campaigns that start themselves, and outcome tracking with review + export. Cloudflare-native and open source (everyai-com/magicbots), with the magicbot fork of OpenMausBot as its sibling.

Why it exists: Work gets discussed in chat, but bots there are toys. MagicBots was built on a simple idea: every bot in the sidebar should be a real agent — its own model, its own computer, its own connected apps — so you can hand whole roles to your team, not just questions.

Features: One bot per role, each a real Claude/Codex/Pi agent; Own model, computer, personality and connected apps per bot; Talk like contacts, watch them work, approve what matters; Voice workspace: AI voice agents + scheduled campaigns; Outcome tracking with review and export; Cloudflare-native and open source — own your data.

How it helps: Delegate entire roles (researcher, caller, writer) to bots that work async while you approve only what matters.

Live: https://bots.magicteams.ai · Repo: https://github.com/everyai-com/magicbots

### AIOS for Founders — Stop being the bottleneck

The founder operating layer: a daily brief, approved follow-through, and one operating system across inboxes, meetings and files — so work keeps moving without the founder in the middle. Vertical workspaces (Home Services, real estate, wealth, email) ride on AIOS Core: tenant policy, durable jobs, a connector gateway (GHL, ServiceTitan, Twilio/Telnyx, calendar) and approval + cost governance with immutable receipts.

Why it exists: Founders become the bottleneck: every inbox, meeting and file routes through them. AIOS was built as the founder operating layer — a daily brief plus approved follow-through — so the business keeps moving without the founder in the middle.

Features: Daily brief + approved follow-through loop; One layer across inboxes, meetings, files; Unified inbox with consents, suppression, approvals, immutable receipts; Connector gateway: GHL, ServiceTitan, Twilio/Telnyx, calendar; Approval + cost governance on agent spend; Vertical workspaces: home services, real estate, wealth, email.

How it helps: Founders stop being the router — work arrives decided or approval-ready, nothing slips, and every action is audited.

Live: https://magicteams.ai

### Sahadeva — Transparent Jyotish workspace · 54 MCP tools

A transparent South Indian Jyotish workspace: Cloudflare Worker engine plus a React mobile web app in English and Telugu. Connector #1 — 54/54 annotation-compliant MCP tools (panchanga chains verified against independent sources), ChatGPT in review, Muse submitted, registry + Smithery live. Public reads; three keyed tools for reviewers.

Why it exists: Jyotish readings are usually opaque — trust me, says the astrologer. Sahadeva was built to make South Indian astrology transparent and verifiable, and to expose it as agent tools so software can work with charts too.

Features: 54 audited MCP tools: panchanga, tithi, muhurta, matching; Verified live against independent almanac sources; React mobile web app in English + Telugu; Public reads; 3 keyed tools for reviewers; ChatGPT in review, Muse submitted, registry + Smithery live.

How it helps: Anyone gets a transparent, checkable chart reading — and developers get Vedic-astrology superpowers inside ChatGPT, Claude and Muse.

Live: https://sahadeva.magicteams.ai · Repo: https://github.com/everyai-com/sahadeva

### Argus — Testing platform

Cloud verification for your software: point it at any web app — deployed or local — and it tests functional flows, UI, UX and visual regressions in parallel on Cloudflare's edge, with feedback looping straight back into your coding agent. A remote MCP service that runs on your existing agent (Claude Code, Codex, Cursor, anything speaking MCP — no API key). Asserts consequences, not appearances.

Why it exists: AI agents ship code that looks done while hiding silent 500s and broken flows — and screenshots can never prove otherwise. Argus was built to verify agent-written software in the cloud, asserting consequences instead of appearances.

Features: Smoke, flows, audits, exploration — deterministic tiers first; Parallel Chromium fleet on Cloudflare edge; Record-once flows with semantic anchors + self-heal; axe-core WCAG, Core Web Vitals, links, visual pixel-diff; Decision envelopes: whatChanged/suggestedFix/nextAction; Remote MCP + CLI + dashboard; cron/CI-able.

How it helps: Ship agent-written code with proof — regressions get caught with file:line fixes before you ever open the app.

Live: null · Repo: https://github.com/everyai-com/argus

### Flare Actions — CI/CD

Open-source GitHub Actions alternative you host on your own Cloudflare. One click deploys the Worker and auto-provisions D1 + queues; GitHub App webhooks flow through edge verification to pull-runners with durable dispatch signals, retry + DLQ and full run history. Warm edge dispatch (no 2–3 min hosted waits), TypeScript + wrangler.jsonc instead of YAML push-test loops, dashboard + API + CLI.

Why it exists: Hosted CI makes you wait minutes in queues and debug YAML by pushing. Flare Actions was built so anyone can own their CI on Cloudflare — warm, fast, typed, and built for agents that verify their own work.

Features: One-click deploy: Worker + auto D1 + queues; Warm edge dispatch, no hosted queue waits; TypeScript pipelines in wrangler.jsonc, local wrangler dev; External pull-runners: labels, containers, services, cache; Dashboard + API + CLI + MCP server; Queue retry + DLQ, full D1 run history, artifacts.

How it helps: Faster, cheaper CI you fully own — and agents can dispatch, wait, read digests and re-run without human babysitting.

Live: null · Repo: https://github.com/everyai-com/flare-actions

### JEV Directory — Evals directory

What the Jev judge-model can do, browsable: 50 runnable evals with exact prompts plus 1,362 real community builds linked to project + source. Agents get a one-paste capability pack, a live MCP server (10 tools + 3 resources) and an llms.txt index. Most-starred repo in the org.

Why it exists: Judge-models are powerful but nobody can see what they can actually do — evals are scattered across repos and chat threads. JEV Directory collects them into one browsable, agent-readable landscape.

Features: 50 runnable evals with exact prompts; 1,362 real community builds linked to project + source; One-paste capability pack for your agent; Live MCP server: 10 tools + 3 resources; llms.txt index for crawlers; Live directory, PR-friendly data files.

How it helps: Pick the right eval in minutes and hand your agent the entire landscape instead of re-explaining it.

Live: https://jev.magicteams.ai · Repo: https://github.com/everyai-com/jev-directory

### Muse Spark Anywhere — Model adapter

Use Meta's Muse Spark in pi, Agent Orchestrator, opencode and any OpenAI-compatible tool — with full reasoning-effort control and no secrets in configs. Collects the small focused fixes for Spark's streaming quirk and discovery gaps per tool.

Why it exists: Meta's Muse Spark is the model behind the muse CLI, but its OpenAI-compatible API has quirks that break strict tools — and catalogs don't list it. This repo collects the small fixes that set it free in every harness.

Features: pi drop-in: models.json + Keychain reader, minimal…max thinking; ~120-line dependency-free SSE proxy for opencode/any GUI; Agent Orchestrator catalog patch for muse-spark-1.3; Draft upstream pi provider; No secrets in configs.

How it helps: Use Muse Spark today in pi, Agent Orchestrator, opencode or anything OpenAI-compatible — with reasoning control.

Live: null · Repo: https://github.com/everyai-com/muse-spark-anywhere

### Leash — Mac menu-bar

Yanks you back the moment your AI agent finishes — or needs you. Hooks finish/needs-input events (Claude Code, Codex, any command), slams a fullscreen overlay with alert + media pause; Enter engages (terminal jumps to front, cursor in prompt), and submitting re-focuses whatever you were doing. A tiny state machine so it only pulls when it actually seized you.

Why it exists: Long-running agents broke the attention loop: you babysit a terminal because you don't know when they'll need you. Leash was built to yank you back the instant they finish — and only then.

Features: Auto-detects Claude Code, Codex, Cursor/VS Code, aider, any command; Fullscreen overlay + alert + media pause on finish/needs-input; Enter engages: agent window jumps to front, cursor in prompt; Auto-returns you to previous work on submit; idle→waiting→engaged state machine, deduped events; Skips the seize when you're already watching.

How it helps: Go full-focus elsewhere while agents run — you'll be pulled back the second you're actually needed.

Live: null · Repo: https://github.com/everyai-com/leash

### Primer — Agent context

The context layer every AI agent is missing. One command maintains live, structured business context connected to your real tools and injects it into any agent automatically — killing the 10–15 minute re-brief every session.

Why it exists: Every agent session starts ignorant, and you burn 10–15 minutes re-briefing it on your business. Primer was built to keep live business context and inject it into any agent with one command.

Features: init wizard scaffolds .primer/ business context; Connectors: files, Todoist, Notion, Gmail, Calendar; sync snapshots fresh data with dates; inject to every harness file: CLAUDE.md, AGENTS.md, cursor, copilot, GEMINI.md; JSON + markdown output for any framework; npx with zero install.

How it helps: Open any agent and it already knows who you are, what you're building, and what happened yesterday.

Live: null · Repo: https://github.com/everyai-com/primer

### AgentProfile — Agent identity

Your agent's identity, everywhere: one profile — skills, credentials, memory — synced to Claude Code, Cursor, Codex and more through a single MCP URL. Open source, zero-knowledge, self-hostable on your own Cloudflare.

Why it exists: Every AI tool starts from zero — its own skills, its own pasted keys, its own amnesia. AgentProfile was built to give all your agents one shared brain through a single MCP URL.

Features: One profile: skills, credentials, memory; Single MCP URL for every client; Zero-knowledge, open source, self-hostable; Live dashboard to watch it work; Works with Claude Code, Cursor, Codex and more.

How it helps: Stop repasting keys and re-teaching every tool — set up once, every agent inherits your setup.

Live: https://agentprofile.everyai-com.workers.dev/app · Repo: https://github.com/everyai-com/agentprofile

### Distillory — Agent memory

Local-first memory engine that synthesizes at ingestion instead of just storing: each source updates one living, schema-graded profile per entity — contradictions resolved, facts grounded in time — so every later read is cheap and already reasoned. One embeddable SQLite file; no server, no Docker.

Why it exists: Most AI memory is a log with an index: it re-derives meaning on every query, expensively. Distillory was built to reason once, at ingestion — so every later read is cheap and already reasoned.

Features: Synthesizes at ingestion into living profiles; Schema-graded, contradiction-resolved, time-grounded; Four verbs: add, search, profile, synthesize (+CLI); Hybrid FTS5 + dense recall with citations; MCP server for any harness; One SQLite file — no server, no Docker, offline.

How it helps: Give agents memory that compounds instead of a pile of logs — local-first and practically free to query.

Live: null · Repo: https://github.com/everyai-com/distillory

### Agent-Ready — Agent gateway

Make your product AI-agent ready: connect your backend, choose exactly what agents may do, click deploy — in under five minutes your product has its own agent gateway (MCP server, API, CLI) running on your infrastructure, safe by default. No API to build, no middleware to write.

Why it exists: Your product works for humans, but agents can't see your data or act for your users — and building that surface takes weeks. Agent-Ready was built to turn any product into an agent gateway in five minutes.

Features: Connect your backend, see it as an agent would; Capability manifest: your signed-off contract; One click deploys MCP server + API + CLI on your account; Safe by default, explicit by choice; Watch an agent actually use your product.

How it helps: Make your product actable by Claude, Codex and every agent — this afternoon, on your own infrastructure.

Live: null · Repo: https://github.com/everyai-com/agent-ready

### Treg — Tool router (fork)

OpenRouter for agent tools: one base URL + one token gives an agent 2,896 catalogued endpoints across 60 providers — SEO, social, enrichment, ads, scraping — priced per call from a cent, no provider signup. Ask for the task, not the tool. Team keys, skills and CLIs callable without credentials ever leaving the server. Self-hostable.

Why it exists: The tools agents need for real work hide behind subscriptions nobody buys for a single run. Treg was built as OpenRouter for tools: ask for the task, not the tool — pay per call, from a cent.

Features: 2,896 endpoints across 60 providers; SEO, backlinks, social, enrichment, ads, scraping; Per-call pricing, no provider signup; Team keys, skills and CLIs callable server-side; Credentials never leave the server; Self-hostable, live at treg.to.

How it helps: Let agents do real research/enrichment/scraping work for cents instead of subscriptions — with safe team sharing.

Live: https://treg.to · Repo: https://github.com/everyai-com/treg

### Relationship Manager — Relationship graph

One graph of the people you actually talk to, with an agent API on top. Reads the graph you already produced — resolving the same human across email, phone, WhatsApp and meeting attendees — kept as evidence-backed records. 17 tools from one contract (remote MCP + REST + CLI), 80 e2e checks against production, deploy-to-Cloudflare.

Why it exists: Your relationships live across Gmail, WhatsApp, calendar and calls — and every CRM demands typing you'll never do. This was built to read the graph you already produced and put an agent API on it.

Features: Resolves one human across email, phone, WhatsApp, meetings; Evidence bands: VERIFIED/PROBABLE, never vibes; prep_brief before every meeting, deterministic and free; Reconnect radar, suppression-respecting; Surfaces: Today, Ask (streaming), People, Pipeline kanban, ⌘K; 17 tools, one contract: remote MCP + REST + CLI.

How it helps: Walk into every meeting prepared and never let a relationship go quiet — the graph does the remembering.

Live: https://relationship-manager.everyai-com.workers.dev · Repo: https://github.com/everyai-com/relationship-manager · Demo: https://www.loom.com/share/2bcc0b3f99e24ecb91c35d4a5410aea7

### LazyStudio — Mac recorder

Record lazily, ship polished videos. Native macOS menu-bar recorder (Swift + ScreenCaptureKit): screen + system audio + mic in one click with a floating camera bubble — then AI cuts silences and retakes, adds zooms, writes chapters and your YouTube title. Auto-detects your installed Claude Code / Codex / Gemini CLI as the editing brain; transcription on-device. Free forever, MIT.

Why it exists: Screen Studio is paid, OBS is complicated, Loom wants a subscription — and nothing open-source does AI auto-editing. LazyStudio was built to fill exactly that gap: record lazily, ship polished.

Features: Native Swift/ScreenCaptureKit menu-bar recorder; Screen + system audio + mic in one click, no drivers; Floating draggable camera bubble; AI polish via your installed Claude/Codex/Gemini CLI; On-device transcription, silence + retake cuts; Auto YouTube title + description next to the mp4.

How it helps: Record in one click and publish a polished video with zero editing skill — free forever, MIT licensed.

Live: null · Repo: https://github.com/everyai-com/lazystudio

### Notch Limits — Usage dashboard

Your Claude Code and Codex limits, living in your Mac's notch. 5-hour + weekly windows, per-model caps, reset times, which account to use next — plus what your usage would cost at API list prices so you can see the deal and run it hot. Ambient dot, hover-to-glance dashboard, installable via Homebrew cask.

Why it exists: Every usage window that resets with headroom is work you paid for and didn't do. Notch Limits was built to show all your Claude/Codex budgets at a glance — so you can run your subscriptions hot.

Features: 5-hour + weekly windows, per-model caps, reset times; Every account polled live in one view; Which account to use next; API-equivalent dollar scoreboard + plan multiple; Ambient notch dot → hover dashboard; Homebrew cask install.

How it helps: See the deal you're getting (~28× your plan cost) and never leave paid tokens on the table.

Live: null · Repo: https://github.com/everyai-com/notch-limits

### Vessel Docs — Docs platform

Open-source, self-hostable documentation layer: Git-backed source compiles to fast human docs, llms.txt context, an MCP-ready surface, generated agent manifest + permission contract + skill.md, drift detection with review-routed fixes — plus a provider-neutral AI assistant with BYOK (OpenRouter, OpenAI, Anthropic).

Why it exists: Docs rot, agents can't consume human docs, and Mintlify-class tooling isn't yours. Vessel was built as the open-source documentation layer for teams building the next interface — for humans and agents alike.

Features: Git source → fast human docs + llms.txt + MCP surface; Generated agent manifest, permission contract, skill.md; Grounded search + read-only docs audit; Drift detection with review-routed fixes; BYOK AI assistant (OpenRouter/OpenAI/Anthropic); Static-first deploy, local dependency-light previews.

How it helps: Ship docs your users love and your agents can actually read — self-hosted, no per-seat docs tax.

Live: null · Repo: https://github.com/everyai-com/mintly-alternative

### PlainSync — Markdown workspace

Markdown for people and AI agents: a calm document editor for humans, exact Markdown for agents. Use the hosted app, install the desktop app (Mac/Windows releases), or self-host the whole stack. Documents stay ordinary .md files.

Why it exists: People and agents need to work on the same documents, but editors lock you in and agents mangle formatting. PlainSync was built so humans get calm and agents get exact Markdown — on plain .md files.

Features: Calm editor for humans, exact Markdown for agents; Hosted app, Mac/Windows desktop, or self-host; Ordinary .md files — import/export freely; Local drafts by default, share when you choose; Comments, roles, history, Yjs collaboration, OpenAPI; MIT — commercial use welcome.

How it helps: One source of truth both you and your agents edit — no lock-in, drafts stay on your device.

Live: https://plainsync-open-markdown.tradephani.chatgpt.site/ · Repo: https://github.com/everyai-com/plainsync

### Muse Field Guide — Field guide

What people actually run with Meta Muse: 351 real use cases, 257 with the exact prompt that ran the job, plus a one-paste capability pack for your agent. Live site fed by a 6-hour sync worker (crowd feed + community PRs + discovery) through KV to the API.

Why it exists: Personal AI agents are new and nobody knows what they're actually good for. This field guide was built from what people really run — 351 cases with the exact prompts — so you can steal what works.

Features: 351 real use cases, 257 with the exact prompt; One-paste capability pack for your agent; Live directory + crowd feed + community PRs; 6-hour sync worker through KV to the API; capabilities.md your agent can read directly.

How it helps: Put your agent to work today with proven prompts instead of discovering everything the hard way.

Live: https://muse.every-ai.com · Repo: https://github.com/everyai-com/muse-use-cases

### Airwallex Kits — Banking kits

Eighteen agentic banking starter kits built on one rule: the model reads, the code decides, the API moves the money. Shared typed REST client, platform/account layer, policy layer, approval gate, sandbox + merchant simulators — 11 kits move real sandbox money, 71 tests green, 12 demo videos.

Why it exists: Moving money with AI is scary without guardrails. These 18 starter kits were built around one rule — the model reads, the code decides, the API moves the money — so agentic finance demos run safely end to end.

Features: 18 kits: typed REST client, policy + approval gates; Platform/connected-account layer; Sandbox + merchant simulators — demos run credential-free; 11 kits move real sandbox money; 71 tests green, 12 demo videos; All 18 kits browsable in a web dashboard.

How it helps: Learn and demo safe agentic banking patterns — approvals first, money movement with receipts.

Live: null · Repo: https://github.com/everyai-com/airwallex-hackathon

### LlamaTutor — AI tutor

Open-source AI personal tutor powered by Llama 3.1 70B via Together.ai inference. Personal one-on-one tutoring for anyone — ask anything, learn at your own pace, free and open source.

Why it exists: Great tutoring is expensive and scarce. LlamaTutor was built to show that open models can teach — a free personal tutor for anyone with a browser.

Features: Llama 3.1 70B from Meta; Together.ai inference; Next.js + Tailwind app; Serper web search grounding; Helicone observability.

How it helps: Learn anything with a patient AI tutor — free, open source, run it yourself.

Live: null · Repo: https://github.com/everyai-com/llamatutor

### MoneyFlow — Trading bot

Polymarket paper-trading bot: local Python scanner + Streamlit dashboard + Cloudflare Worker snapshot API with scheduled refreshes. Scans live markets every 2 minutes, computes edge from price + orderbook imbalance, filters by volume, liquidity, time-to-resolution and question type.

Why it exists: Prediction markets move in minutes and manual trading can't keep up. MoneyFlow was built to scan Polymarket continuously, compute real edge, and paper-trade it — with full P&L honesty.

Features: Live scans every 2 minutes; Edge from price + orderbook imbalance; Volume, liquidity, time + type filters; Kelly Criterion bet sizing; Google Sheets log + email alerts on every signal; P&L tracking with auto 12-hour reports; Streamlit dashboard + Worker desk UI + Exa briefs.

How it helps: Trade predictions with quantified edge on autopilot in simulation — learn what works before risking a cent.

Live: null · Repo: https://github.com/everyai-com/MoneyFlow

### Sahadeva Assets — Design system

Source-controlled visual language for Jyotisha products — traditional in meaning, modern in expression: 156 visual concepts (9 navagraha, 12 rasi + 12 bhava, 27 nakshatra, 30 tithi…) plus brand marks, lockups, app icons and production exports.

Why it exists: Jyotisha products need visuals that are traditional in meaning but modern in expression — and everyone redraws them inconsistently. This library was built as the source-controlled visual language for the whole family.

Features: 156 concepts: navagraha, rasi, bhava, nakshatra, tithi, yoga…; Brand marks, lockups, app icons, exports; Semantic light/dark tokens, type, motion, a11y states; Reference React components + screen studies; npm CSS imports, browsable catalogue.

How it helps: Ship authentic, accessible, consistent astro UI in days instead of designing every symbol from scratch.

Live: null · Repo: https://github.com/everyai-com/sahadeva-asset-library

### Velocity — Reading app

A premium speed-reading studio: load any book, read at the speed of thought. Personal shelf, hosted on Cloudflare Pages in our account.

Why it exists: Reading speed gates how fast you learn, but speed-reading tools feel like toys. Velocity was built as a premium studio: pin your eyes to one point and read at the speed of thought.

Features: RSVP-style fixed-point reading; EPUB, PDF and TXT uploads + drag & drop; Quick-read: paste any text instantly; Personal shelf + reading stats; Free Gutenberg classics built in; Hosted on Cloudflare Pages.

How it helps: Get through books and long reads multiples faster — your whole library with stats on your shelf.

Live: https://velocity-reader-6oa.pages.dev

### Uniqu — Accessibility

Honest web accessibility: WCAG/ADA/EAA compliance scanning, an embeddable one-file widget, and platform-aware fix guidance with publishable evidence. $299 once — deterministic rules, no AI theater.

Why it exists: Accessibility overlay widgets overpromise while ADA/EAA lawsuits keep rising. Uniqu was built for the honest path: find real failures, fix them at the source, and publish evidence you can stand behind.

Features: URL → WCAG issues via Playwright + axe-core; Embeddable widget ships as one JS file; WCAG + custom rule definitions; Platform-aware fix guidance (Shopify-first); Evidence reports you can publish; $299 once — deterministic rules, no AI.

How it helps: Online businesses fix real barriers and prove the work — lawsuit protection you can show, not a badge you hope holds.

Live: https://uniqu-marketing.pages.dev

### Reroof Solar — Solar sales

Commercial solar proposal machine: finds aging-roof buildings, pulls real panel layouts via Google Solar API, generates personalized prospect microsites + Claude-written outreach, and tracks everything on a pipeline dashboard.

Why it exists: Solar companies prospect by hand and pitch generic numbers. Reroof was built to automate the whole commercial pipeline — from finding aging roofs to booked appointments — with real building-specific solar data.

Features: Sources buildings via Google Places/Regrid; Real layouts + output + savings via Google Solar API; Personalized microsite per prospect; Claude-written personalized outreach; Pipeline dashboard: new → processed → emailed → replied → booked; Supabase + React + Vercel + n8n.

How it helps: Turn prospecting into a machine: every outreach carries that building's real numbers, tracked to booked.

Live: null

## MCP Connectors (26)

Build once, listed in ChatGPT, Claude and Muse directories. All workers verified live.

- Sahadeva (vedic-astrology, submitted, 54 tools): Vedic astrology engine: panchanga, tithi charts, muhurta search. 54/54 annotation-compliant tools, verified live against independent sources. Connector #1. — MCP: https://sahadeva.magicteams.ai/mcp — ChatGPT: in-review, Claude: needs-login, Muse: submitted
- FestivalFinder (festivals, submitted, 4 tools): Computed Hindu festival dates 2020–2040 from tithi rules + sankrantis. All read-only, validated within ±1 day of independent 2026 calendars. — MCP: https://festival-finder.magicteams.ai/mcp — ChatGPT: in-review, Claude: needs-login, Muse: submitted
- InvoiceGen (invoicing, submitted, 4 tools): Freelancer invoices: numbering, totals, tax, text render. Pure read-only compute, nothing stored — math verified live ($531 test). — MCP: https://invoice-gen.magicteams.ai/mcp — ChatGPT: in-review, Claude: needs-login, Muse: submitted
- ReportRaja (reports, submitted, 4 tools): Weekly status reports: templates, week bounds, text render. Registry + Smithery live; ChatGPT in review. — MCP: https://report-raja.magicteams.ai/mcp — ChatGPT: in-review, Claude: needs-login, Muse: submitted
- SheetShift (spreadsheets, deployed, 4 tools): Spreadsheet surgery: convert tables, clean data, column stats, split columns. ChatGPT draft in review (verified by pipeline owner). — MCP: https://sheet-shift.magicteams.ai/mcp — ChatGPT: in-review, Claude: needs-login, Muse: pending
- FlightFix (travel, deployed, 5 tools): EU261/UK261 flight compensation: eligibility checks, rights briefs, claim timelines, generated claim letters. ~170-airport table. — MCP: https://flight-fix.magicteams.ai/mcp — ChatGPT: draft, Claude: pending, Muse: submitted
- CourtSplit (sports, deployed, 4 tools): Court-cost splitting with cents-exact math, minimal-transfer settlement, recurring-session accounting and fair rotation scheduler. — MCP: https://court-split.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- LeaseBreak (rentals, deployed, 4 tools): Lease-break math: itemised break costs, break/sublet/stay comparison, notice letters, input-tailored negotiation checklists. — MCP: https://lease-break.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- AIReady (ai-visibility, deployed, 5 tools): AI-crawler readiness (GEO): site scans plus robots.txt / llms.txt / schema generators across a 13-bot crawler table. The “get visible with AI” play. — MCP: https://ai-ready.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- LeagueNight (sports, deployed, 4 tools): Sports league toolkit: circle-method round-robin with byes + second legs, standings, courts-by-slots match-day grids, season sizing with dates. — MCP: https://league-night.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- FeeFighter (personal-finance, deployed, 4 tools): Fee audits with cents-exact annualisation, 1–30y projections, 14-category US fee benchmarks and dispute letters with escalation paths. — MCP: https://fee-fighter.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- RoomSplit (housing, deployed, 4 tools): Roommate money: rent splits by room size or equal shares, utility splits, minimal-transfer settle-up, plain-language living agreements. — MCP: https://room-split.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- PartyPlan (events, deployed, 4 tools): Party planning: cents-exact budgets in three styles, headcount-to-quantity plans with buffers, T-42 countdown timelines, home-or-venue checklists. — MCP: https://party-plan.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- ServiceSchedule (automotive, deployed, 4 tools): Car maintenance planning: due services vs typical km/month intervals, US cost ranges by shop tier, 12-month projection, seasonal checklists. — MCP: https://service-schedule.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- RecipeScale (kitchen, deployed, 4 tools): Kitchen math: serving scaling with kitchen-fraction hints, volume/weight conversion via ingredient densities, merged shopping lists, cost per serving. — MCP: https://recipe-scale.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: submitted
- TutorNow (family-food-health, deployed, 7 tools): Tutor booking — first product with writes. 7 tools with per-tool read/write/destructive annotations, the template pattern for future write tools. — MCP: https://tutor-now.magicteams.ai/mcp — ChatGPT: packet-ready, Claude: packet-ready, Muse: packet-ready
- GroceryStock (family-food-health, deployed): Grocery inventory tracking connector (idea FF06). Worker live, demo rendered; submission packets in the factory pipeline. — MCP: https://grocery-stock.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- ClassDrop (sport-social-events, deployed): Class drop-in booking connector (idea SS06). Worker live, demo rendered; submission packets in the factory pipeline. — MCP: https://class-drop.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- SalonBook (local-recurring, deployed): Salon appointment booking connector (idea LR01). Worker live, demo rendered; submission packets in the factory pipeline. — MCP: https://salon-book.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- MowRecur (local-recurring, scaffolded): Recurring lawn-care scheduling connector (idea LR07). Scaffolded 2026-10-05; tools being implemented. — MCP: https://mow-recur.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- PlumbNow (home-urgent, deployed): Urgent plumber dispatch connector (idea HU01). Deployed 2026-10-05; directory ship queued. — MCP: https://plumb-now.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- HvacRescue (home-urgent, deployed): HVAC rescue dispatch connector (idea HU04). Deployed 2026-10-05; directory ship queued. — MCP: https://hvacrescue.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- PestBlitz (home-urgent, deployed): Pest-control dispatch connector (idea HU07). Deployed 2026-10-05; directory ship queued. — MCP: https://pest-blitz.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- MealKitMatch (family-food-health, deployed): Meal-kit matching connector (idea FF05). Deployed 2026-10-05; directory ship queued. — MCP: https://meal-kit-match.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- PadelCourt (sport-social-events, deployed): Padel court finder + booking connector (idea SS01). Deployed 2026-10-05; directory ship queued. — MCP: https://padel-court.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending
- StayTonight (travel-stay-move, deployed): Same-night stay finder connector (idea TS01, renamed from hotel-tonight). Deployed 2026-10-05; directory ship queued. — MCP: https://stay-tonight.magicteams.ai/mcp — ChatGPT: pending, Claude: pending, Muse: pending

## Revenue Agents (3)

One shared core (agents/_core v0.1: worker + D1 + owner console + Stripe + approvals/audit). Staging: https://agents-core.everyai-com.workers.dev

### MissedCallBack — Missed call → text-back → booked job

Missed call → SMS text-back in seconds → qualification (job → address → urgency → window) → booking link → owner alert. Emergencies (gas, no water, flooding, sewage) escalate to the owner's cell immediately. Full TCPA rails: consent copy on first message, STOP/HELP handling, suppression list, quiet hours with cron flush. Every action lands in the audit log with a consent snapshot. Flow built in agents/_core v0.1, local e2e green; awaiting vendor credentials to go live.

### VoiceReceptionist — 24/7 AI answering on the MagicTeams platform

Voice answering runs on the existing MagicTeams platform (Ultravox + Twilio/Telnyx, inbound agent per number) — reuse, don't rebuild. This product adds missed-call text-back, SMS qualification, owner alerts and an mtCreateAgent client that provisions the receptionist (name + system prompt from per-vertical greeting/transfer rules). Shared core built; voice wiring is the remaining vendor step.

### ReviewResponder — Never miss a review again

New review → conversation + on-brand drafted reply → owner approval required (approve to post, reject to close; both audited). 1–2★ reviews escalate to the owner's cell before anything else happens. Monday 09:00 local digest covers new reviews, average rating and pending drafts. Drafts are banded by rating, never offer incentives, never gate reviews, and always route unhappy customers to a direct conversation. Flow built in core, e2e green; live posting waits on GBP API access.

## Distribution

- ChatGPT Plugins: 5 in review — sahadeva, festival-finder, invoice-gen, report-raja, sheet-shift — sidebar homes + MCP-events automations audience.
- Claude Directory: 5 packets ready — Blocked on portal login (any paid plan submits). Community connector auto-scan path.
- Meta Muse: 11 submitted — Newest directory (~37 live connectors). Report-raja + sheet-shift next in portal.
- Smithery: 20/26 live — tradephani/<slug> via CLI. CI publishes via tags once repo secrets land.
- MCP Registry: 2/26 live — invoice-gen + recipe-scale published. CI publish queued for the rest.
- xAI Grok: PR open — xai-org/plugin-marketplace#1206 — one magicteams-connectors plugin, validator green.

## Operator

MagicTeams / EveryAI — support@magicteams.ai — https://magicteams.ai — https://github.com/everyai-com