recast

01 / recast mcp · access release

everyone just got the same genius. we sell the difference.

every new model makes every ai smarter — and nobody's ai different. recast mcp is the knowledge layer that plugs 200 assets across 13 categories, distilled from years of agency work, into claude or cursor through one url. your ai searches it on every task, on this model and the next.

for operators & builders already working in claude or cursor

Join the waitlist

name + whatsapp · we message you the moment access opens

the 2-minute connection
mcp · open standard

claude mcp add recast https://mcp.recast-agency.com/c/‹your-key›/mcp

connected · recast mcp is live in this session

200 assets across 13 categories · semantic search on every task

weekly drops sync to the same url — nothing to reinstall

‹your-key› is issued on your access call

illustrative

200
assets in the corpus
13
categories
2 min
to connect
weekly
new drops

02 / the shift

The model stopped being the edge.

every frontier model ships to you and your competitor on the same morning. fable 5 is just the latest. when intelligence is evenly distributed, output converges — the edge moves to what your ai knows that everyone else's doesn't.

the convergence · one model, run bare vs. run through the brainillustrative
youyour competitorthe model=

same model · same page

two businesses prompt the same model. it returns the same page — twice.

youyour competitorrecast brainthe model≠

recast brain · 200 assets · 13 categories · semantic search

your competitor prompts the model. your prompt searches your corpus first.

same prompt · same model — the only variable is the corpus your ai searches first

01 · same genius

your competitor's claude is exactly as smart as yours.

intelligence is rented now, and everyone rents the same model. the ceiling is shared — you can't out-buy it, and you can't out-prompt it for long.

02 · average by default

an empty model writes the internet's average.

given nothing of yours to read, it reaches for the most common pattern it knows. that's the first board — same structure, same copy, same page, twice.

03 · where the edge went

the edge is the corpus your ai searches first.

200 assets across 13 categories, searched by meaning before a line gets written. that layer can't be prompted into existence — it has to be built.

03 / under the hood

A brain your AI can search.

MCP is the USB port of AI — an open standard that plugs outside knowledge straight into Claude or Cursor through one URL. Behind ours sits a live server: 200 assets across 13 categories, searched by meaning — not keyword — with new drops landing weekly. Most pages would stop there. Here's the dissection.

200 assets · 13 categories · searched by meaning · weekly drops

The Recast brain — a semantic knowledge graph of luminous nodes and edges on black, one cluster ignited molten orange where a search matched
the brain · 200 assets as a semantic graph — searched by meaning, not keywords.

the connection · one url, every model

mcp · the usb port of ai

Diagram: Claude, Cursor, and whatever MCP client ships next quarter all connect through one URL into the recast brain — 200 assets across 13 categories — and from the brain one line continues into your output, unmistakably yours. Models rotate; the corpus stays.

one url · an open standard · models rotate, the corpus stays

exhibit a · the corpus

all 13 shelves · ranked by depth

A vector-database index — thousands of glass data cards in an isometric grid; a wave of orange light sweeps through, igniting the cards a search matched.
the index — a search lights up the cards that match.

  • websites

    109

    the deepest shelf — page patterns, templates, build systems.

    • ws-117Website design craft — anti-AI-look rules
    • ws-029Website build prompt: conversion landing page

  • marketing

    22
    • mk-063Landing page copy sequence
    • mk-061Competitive messaging map

  • ugc-video

    21
    • uv-077UGC writer output schema
    • uv-085Beat chaining for short AI ads

  • ui-patterns

    15
    • up-101AI input composer pattern
    • up-102Tooltip and command surface pattern

  • prompting

    7
    • pr-006Reusable prompt-variable system
    • pr-002Character consistency across angles

  • copywriting

    6
    • cw-011Above-the-fold landing copy formula
    • cw-118VSL script — the 9-beat flow

  • coding

    4
    • cd-021Bug repro-first prompt
    • cd-019Refactor-to-remove-duplication prompt

  • business

    3
    • bz-026From skills to first paying client
    • bz-027Positioning an AI offer

  • design

    3
    • ds-016Landing page hero critique checklist
    • ds-017Design review rubric

  • ugc

    3
    • ug-015Creating UGC ads with AI — pipeline
    • ug-014AI talking-head video workflow

  • video-editing

    3
    • ve-124AI video editing with Claude Code — system overview
    • ve-125Multi-take shoot → final cut — the EDL playbook

  • automation

    2
    • au-024Data-entry agent guardrails
    • au-023Notion tidy-and-format agent spec

  • operations

    2
    • op-112Video analysis workflow
    • op-113Vision model anti-hallucination guardrails

200 assets · 13 categories

new drops land weekly

exhibit b · the four tools

the entire api surface

ping

the handshake — confirms the server is live, who you are and what plan you're on.

returns status · tenant · plan

list_categories

the shelf map — every category you can access, with live counts.

13 categories today

fires on every task

search_assets

semantic, meaning-based — not keyword. returns ranked snippets with relevance scores.

optional category filter · up to 20 hits

get_asset

the pull — fetches the complete body by id. search returns previews only.

one id in · the full asset out

exhibit c · the working loop

three calls per task

Diagram: the working loop — search a topic with search_assets, review the ranked snippets it returns, then pull the complete asset by ID with get_asset.

exhibit d · specimens

real assets · illustrative scores

competitive-messaging-mapmarketing

Competitive messaging map

analyze competitor copy and find the angle nobody owns.

relevance0.92
brand-memory-systemmarketing

Brand memory system

marketing agents that compound knowledge over time.

relevance0.89
conversion-diagnosis-treemarketing

Conversion diagnosis tree

walk the funnel and find the exact breakpoint.

relevance0.87
vsls-vs-webinars-vs-sales-deckscopywriting

VSLs vs webinars vs sales decks

the deep-dive on choosing between the three long-form sellers.

relevance0.81

Not a course. Courses get watched once. This fires on every task, inside the tool you already use.

Join the waitlist

name + whatsapp · we message you the moment access opens

04 / the demo

Watch Claude think with your brain plugged in.

same chat you already use. one prompt goes in — claude searches the corpus over meaning, not keywords, pulls the frame that fits, and builds from it. this is that conversation, beat for beat.

Claude
Fable 5zuheir-demo · partner

build the landing page for a med-spa in austin

Let me search the recast corpus for a frame that has already converted for this.

Building from ws-041: pattern-interrupt hero, proof-stacked offer, 2-step booking flow.

This page starts where the average page ends.

recast mcp · 200 assets · 13 categoriesillustrative — the live demo happens on your call

the model alone

the internet’s average med-spa page — same hero, same three cards, same cta as everyone else’s.

the model + recast

the same model building from a frame that already sold — structure the average page never reaches.

see it on your business

Join the waitlist

name + whatsapp · we message you the moment access opens

05 / how it works

Connect once. It compounds.

No new tool, no migration — it plugs into the AI you already use, and the brain underneath keeps growing.

01

connect one URL

your key, your seat, two minutes

02

your AI searches the brain on every task

pages, ads, funnels, video — built from proven systems

03

new mechanics ship weekly

your output upgrades while you sleep

macro footage: a fiber-optic cable plugged into a matte-black server port, orange light pulses traveling down the line
one url. that's the whole integration.
works withclaude codeclaude.aicursormodel-agnostic · open standard

200 assets in the corpus · 13 categories · 2 min to connect · weekly new drops

06 / the drops

The brain doesn't wait for the next model.

New assets land weekly through the same URL — nothing to reinstall, nothing to migrate. Whoever ships the next model, your corpus is already deeper than it was last week.

The drop cadence: three drops already merged into the corpus, and this week's drop already live in your tool.

  1. drop · n-2

    landed · in the corpus

  2. drop · n-1

    landed · in the corpus

  3. drop · last week

    landed · in the corpus

  4. An obsidian glass data capsule with a molten orange core — this week's drop, sealed and already merged into the corpus
    this week · already in your tool

200 assets today · deeper every week · same url

Join the waitlist

07 / asked every time

Same four questions, every call.

answered here the way we answer them on the phone — so the call can be about your business, not ours.

01

can't I build my own MCP server?

Sure — the protocol is open. What you can't build in a weekend is the corpus: 200 assets across 13 categories, distilled from live builds and updated weekly. The server is plumbing; the brain is the product.

02

does it work with my stack?

Claude Code, claude.ai and Cursor today. Open standard — the surface grows on its own.

03

doesn't the newest model make this unnecessary?

Opposite. Every release — Fable 5 included — raises everyone's floor; the brain raises your ceiling. Rising tide, same boats — unless your boat has an engine.

04

what happens when the next model ships?

Your brain comes with you. MCP is an open standard — the same URL plugs into Claude, Cursor and whatever ships next quarter. Models rotate; the corpus compounds.

you have the same genius as everyone else. for now.

fable 5 is free on pro, max and team until june 22. connect while everyone else is still marveling at the model — the window closes, the brain doesn't.

Join the waitlist

name + whatsapp · we message you the moment access opens

what happens next

01 · drop your name + whatsapp02 · we hold your spot in the next batch03 · we message you the moment access opens

early access opens in limited batches.