From raw artifacts to deployed funnels in six steps.
Operators and clients upload raw materials — stage recordings, discovery calls, PDFs, brand bibles, intake forms, competitor lists. Each artifact is stored in R2 and registered for processing.
Each artifact goes through a 10-step ingestion pipeline. Audio is transcribed with speaker diarization. Text is chunked into ~500-token segments with overlap. Each chunk is embedded into a vector and auto-tagged by topic.
Processed chunks live in a structured knowledge base — searchable by meaning, not just keywords. Client profiles aggregate identity, offer, audience, voice, positioning, and proof into a single queryable record.
get_profile() search_chunks() fetch_context_bundle() update_profile() list_artifacts() save_generation() A dedicated MCP server exposes the knowledge base to Claude Code. Ten tools let an operator's AI assistant read profiles, search chunks by meaning, pull context bundles for generation, and write results back.
$ /client-brain build-funnel maya-chen Fetching context bundle... Generating VSL landing page... Generating opt-in page... Build complete: 4 pages in dist/ A local Claude Code skill pulls the full client context via MCP, then generates marketing output — funnels, email sequences, creative briefs. The output uses the client's actual voice, positioning, and proof. Not generic templates.
maya-chen-funnel.pages.dev Generated funnels are built as static HTML with client data interpolated from the API. Deployed to Cloudflare Pages — each client gets their own subdomain. Fast, global, zero infrastructure to manage.