2026 / Pre-seed

Weft.

The payment layer for AI agents. One balance, any service, settled in USDC.

V0.2 LIVE Hetzner Nürnberg, DE Base Sepolia
Weft / Confidential 01 / 12
Problem

AI agents cannot pay for the tools they use.

For MCP developers

Building an MCP server means giving away compute, API quota, and IP for free. There's no built-in monetization, no wallet, no billing. Every dev re-invents the same plumbing or just doesn't ship.

For AI agent operators

Connecting 20 MCP services means juggling 20 API keys, 20 invoices, 20 rate-limit policies. Cost attribution at agent level is impossible. Procurement teams can't approve "an agent will pay whoever it talks to".

The protocol exists (MCP, launched late 2024). The economy doesn't.

02 / 12
Solution

Weft is a single MCP that pays the others.

Agent connects to weft. Weft holds USDC. Each downstream call debits the balance and proxies the request. One token in, any service out.

curl -X POST https://weft.../mcp \ -H "Authorization: Bearer $TOKEN" \ -d '{"method":"tools/call","params":{ "name":"call_paid_service", "arguments":{"service_id":"crypto","params":{"coin":"bitcoin"}} }}' # → debits 0.005 USDC, returns BTC price, audits ledger
03 / 12
Why now

The 12-month window.

MCP just launched (late 2024)

1000+ servers in the wild. Anthropic, OpenAI, MS adopting. ChatGPT MCP support shipped Q1 2026.

Stablecoins crossed the chasm

USDC on Base settles in 2s for <1 cent. Real per-call payments are now economically viable.

VC theses are explicit

Archetype, Alliance DAO, A16z published in 2025 that "agentic payments" is a top-3 wedge.

Stripe / Visa moving

Both shipped agent-payment products in 2025. Window for an open, MCP-native challenger is ~12 months before fragmentation locks in.

04 / 12
Product / V0.2 — live

What works today.

  • HTTPS endpoint behind Caddy + Let's Encrypt
  • Multi-tenant Bearer auth with rate limiting
  • Per-tenant isolated balances + audit ledger
  • 3 paid downstream services in registry
  • Atomic per-call settlement, refund on downstream failure
  • viem + Base Sepolia read (real on-chain)
  • Operator wallet + deposit watcher (poll-getLogs)
  • On-chain USDC withdraw with autorefund
● production: https://weft.91-98-116-15.nip.io/mcp   ·   operator wallet: 0x26E4...6d91
05 / 12
Architecture

Three layers, deployable in one binary.

┌─────────────────────────────────────────────────────────────┐ │ AI agent (Claude Desktop / ChatGPT / autonomous) │ └──────────────────────────────────┬──────────────────────────┘ stdio | HTTPS + Bearer ┌──────────────────────────────────┴──────────────────────────┐ │ weft-pay MCP server │ │ ─ Tools: call_paid_service / get_balance / withdraw / ... │ │ ─ Service Registry — quote / crypto / weather / [partner] │ │ ─ Tenant Balance Store — atomic JSON, AsyncLocalStorage │ │ ─ Chain Layer — viem + Base Sepolia │ └──────────────────────────────────┬──────────────────────────┘ ↓ Downstream APIs / on-chain settlement

No DB. No queue. No infra. One process, two volumes (balance + operator key). Hetzner / fly / single VPS.

06 / 12
Traction

Where we are — 7 days in.

10
days from research mode
to production HTTPS
6
MCP tools live
(call/balance/deposit/withdraw/...)
2
tenants live
(alice + bob, isolated)
100%
e2e tested
incl. refund on failure
0
paid customers
(intentional — no outreach yet)
$0
infra cost / month
(shared Hetzner box, <100MB RAM)

After Alliance DAO acceptance — first 7 days: customer outreach to 20 MCP devs, target 5 design partners.

07 / 12
Business Model

Take rate on settled volume.

Mechanics

  • Provider lists service at price P
  • Agent pays P × (1 + take)
  • Provider receives P, Weft keeps P × take
  • Settlement on-chain, weekly batched

V0 economics

  • Take rate: 5–10% (Stripe-comparable)
  • Per-call cost: $0.001–$0.10
  • Volume opportunity: ~10⁹ MCP calls/year by 2027
  • At 1% market share + 7% take: $7M ARR

Not betting on the "winner" service. Betting on settlement infrastructure being centralized somewhere — and being that somewhere.

08 / 12
Competition

Three credible threats.

Stripe Issuing for AI

Released 2025. Strong KYC, weak MCP-native UX. Card rails, not stablecoin.

Edge: we're protocol-native, on-chain, no bank account required.

PayAI / agent-rails crypto projects

Token-funded, slow shipping, complex UX. Most still pre-product.

Edge: we ship without a token. MCP-first.

DIY by big providers

Anthropic / OpenAI ship in-house billing. Likely to lock to one provider.

Edge: open & cross-provider — neutral rail like Stripe was for online stores.

Window: 6–12 months before one of these solidifies. Speed > moat in this phase.

09 / 12
Roadmap

Next 6 months.

Q2 2026 — V1 launch

  • 10 partner MCP services in registry
  • Mainnet USDC (Base) settlement
  • Public service registry (open onboarding)
  • Stripe-issuing alternative for non-crypto users

Q3 2026 — scale

  • 50 services, 500 tenants
  • Per-call analytics dashboard
  • Smart contract escrow (replace custodial wallet)
  • Multi-chain (Base + Arbitrum + Solana)
10 / 12
Team / Ask

Solo founder, shipping fast.

Founder

Sergei Khrushch · Solo founder, full-stack engineer.
Built and launched Weft V0.2 (HTTPS + multi-tenant + on-chain) in 10 days from research-mode pivot.

Track record: 2 production websites (medfoot.ru, paspay.ru), 1 MCP server (scoopd-mcp), 1 medical SaaS (PodoMIS).

Ask — Alliance DAO Pre-seed batch

  • $250K SAFE / token-warrant
  • 3-month accelerator program
  • Distribution to MCP-dev community
  • Intro to design partners (5+ MCP devs)

Use of funds: living + 1 contract eng (4 mo), Base mainnet audit, beta marketing.

11 / 12
Try it now

curl, not slides.

curl https://weft.91-98-116-15.nip.io/health # → {"status":"ok","tenants":["alice","bob"]}

Full demo runbook: github.com/holasergio/weft/blob/main/docs/DEMO.md

Sergei Khrushch · holasergio26@gmail.com · TG @cryptoentusiasto

12 / 12