An AI infrastructure stack, built entirely from a phone.
Persistent memory for Claude. Web hosting you drive from a chat window. Full automation control. Each one is a custom MCP server I designed, coded, and deployed without ever opening a laptop — and this page itself was built and put live the same way.
every node is a real captured memory · drag, zoom, hover
0
0 relations and counting
The three servers
All three run as Cloudflare Workers and plug straight into Claude — web, mobile, Claude Code, and Cowork — through the Model Context Protocol.
Persistent AI memory
Second Brain MCP
Claude forgets everything between chats. Second Brain doesn’t. Every substantive conversation — across web, mobile, Claude Code, and Cowork — is captured, compressed, and linked into the knowledge graph you just watched build itself. The next session starts by pulling a tiny context payload instead of re-explaining anything, so it saves tokens instead of burning them.
It remembers every project, person, decision, and open thread ever discussed — and connects them. Ask it anything about your own history and it answers from evidence, with hybrid recall that fuses semantic vectors, keyword search, entity facts, and graph relations in parallel.
23 toolsFrom instant key-value lookups to full natural-language Q&A over the whole graph.
Hybrid recallVector + keyword + entity + relation search fused with rank aggregation and reranking.
Session handoffsEvery session ends with a handoff note; the next one resumes exactly where you left off.
Token-efficient by designCompressed wiki entries instead of raw transcripts — roughly 27× smaller to recall.
A hosting pipeline that lives inside a conversation. Describe a site, and Claude builds it, deploys it to Cloudflare’s edge, and hands back a live URL — in one message, from a phone. No terminal, no dashboard, no computer. It’s the difference between “I’ll send you a mockup next week” and “here’s the live link, refresh now.”
The page you are reading was created, deployed, and shared exactly this way.
Deploy from chatHTML or multi-file bundles go straight to a live *.pages.dev URL. Projects auto-create.
Instant rollbackEvery deployment is versioned; promote any previous one back to production.
Inspect live sitesFetch any deployed file back into the chat to review and edit in place.
Real client workDemo sites, proposals, and landing pages shipped to real prospects — see below.
Full command of an n8n automation instance from inside Claude. Create workflows, patch them, validate them structurally before they run, flip triggers on and off, fire webhooks, and read execution logs — the entire lifecycle, conversationally. An OAuth Worker shim makes it available to Claude on every surface, doing everything the official integration does and more.
Full CRUDCreate, read, update, patch, and delete workflows and credentials by chat.
Validate before runStructural workflow validation without triggering anything — catch breaks early.
Execution visibilityList and inspect runs, filter by workflow, delete records — debugging without a dashboard.
Everywhere Claude isOAuth-MCP shim on Workers: works in Claude web, mobile, Code, and Cowork.
Cloudflare Workers OAuthn8n RESTWebhook execution
Why this exists
I’m disabled. I work from bed, from a phone. So I built the tools that make that irrelevant.
Most developer workflows quietly assume a desk, a laptop, and a body that cooperates. Mine doesn’t. Instead of accepting a smaller output, I rebuilt the stack around my constraints: memory that persists so nothing has to be re-explained, hosting that ships from a chat window, automations that run themselves and report back.
The result isn’t a workaround — it’s a faster pipeline than most people have with a full desk setup. Everything on this page, including this page, was designed, built, deployed, and delivered from a phone.
Shipped with this stack
A sample of live deployments produced through the Pages MCP — demos, proposals, and client-facing sites, all built and hosted the same way.