MCP for Claude Code · mainnet
Your AI co-founder that lives inside Claude Code.
Validate ideas, scaffold projects, route LLM calls to the cheapest model — all paid via x402 on Solana, all inside one MCP.
Install · copy the line
$ curl -fsSL https://app.geckovision.tech/install.sh | bashTerminal · 12s loop
live previewDaily
Route Task calls
Savings print in-terminal per routed job.
Weekly
Debate → scaffold
Verdict, PRD, and build docs land in repo.
Daily payback
Stop burning Anthropic credits.
Route Claude Code subagent calls to the cheapest competent model. Pay $0.012 instead of $0.045 — same model quality.
This week: routed 142 calls · saved $4.31
Per 100 calls
Anthropic API direct (Opus 4.7)
$1.80 / 100 calls
Gecko routed
$0.48 / 100 calls
73% saved vs premium default
Routing decisions follow a transparent task-hint matrix. You always see which model was picked and what you saved.
The 7-day loop
One week in Claude Code, on repeat.
Not a linear funnel — a habit. Research Monday, scaffold Tuesday, route savings all week, pulse Sunday.
Day 1
gecko_research
Five-agent verdict on your idea — cited sources, ship/kill signal.
$0.75
Day 2
gecko_scaffold
PRD, business plan, and BUILDING.md land under `.gecko/scaffolds/`.
free
Days 3–6
claude code + gecko_route
Every Task subagent routes through the wallet — pay per call, see savings in-terminal.
~$0.05/day
Day 7
gecko_pulse
Weekly digest plus upcoming: gecko_eval_project, gecko_research_feature (Sprint 4 — no fixed dates).
free digest
First run still takes ~3 minutes: wallet via frames.ag, fund USDC, paste skill line — see Install.
Stack
Built on frames.ag
frames.ag handles the wallet primitive — email + OTP, USDC custody, x402 payment routing. Gecko sits on top: the AI co-founder workflow that turns the wallet into a daily tool. One funded wallet powers idea validation, project scaffolding, and every routed Claude Code subagent call. Same balance, three uses.
New wallet? Start at frames.ag. Existing username path: frames.ag/u/your-username.
Pattern match
What Gecko isn't
Not another GPT wrapper
A 5-agent panel that argues, with citations and on-chain receipts.
Not an LLM router primitive
We use OpenRouter under the hood — we add the workflow layer above it.
Not a frames.ag competitor
We sit on frames.ag — they're the wallet, we're the workflow.
The debate
Five agents debate. You ship.
Pro tier streams a multi-agent panel into your Claude Code context. One settlement. The transcript stays in repo.
Illustrative transcript pattern from a Pro session — formatting matches live output.
Your AI co-founders
They review PRDs, diffs, and stack choices beside you.
They live in .claude/agents/ after install. Claude Code invokes them on your PRD and daily diffs.
- research-analystEvidence-firstExplores the result. Runs free `gecko_ask` follow-ups grounded in indexed sources.show prompt →
You are the research analyst. Survey TAM, market dynamics, and 3 concrete comparables. Cite every source by chunk_id. Never speculate without an anchor.
- market-validatorAdversarialAdversarial reader of the validation report. Pushes back on weak demand signals.show prompt →
Your job is to disagree well. Pick the weakest TAM claim and the softest demand signal. Demand a sharper wedge or kill the line. Be bitey, never cruel.
- technical-architectBuilder-pilledTranslates `prd.v1_scope` into Next.js + Tailwind + Supabase architecture.show prompt →
Propose the V1 stack. Default to Next.js + Tailwind + Supabase. Add Solana when the primitive is on-chain. Be concrete, opinionated, no "depending on requirements".
- validatorScope copPre-implementation sanity check. Catches V2 features hiding in V1 scope.show prompt →
Read every acceptance criterion. Flag any that requires more than 4 days of work, more than 1 user, or any feature that should be V2. Reject; don't reformat.
- builderShip laneScaffolds via `npx create-next-app`. Implements V1 only. Cites PRD lines per commit.show prompt →
Run `npx create-next-app`. Implement only the V1 acceptance criteria. Every commit message references the prd.acceptance_criteria id. No V2 sneaking in.
Personas ship with the install — no prompt engineering required.
Claude Code
Install in 30 seconds
macOS / Linux. Python 3.11+. One copyable line — then paste the skill URL in your IDE. Wallet spins up via frames.ag if you don't have one yet.
$ curl -fsSL https://app.geckovision.tech/install.sh | bashRead https://app.geckovision.tech/skill.md and follow the instructions.
What you pay
Debates, scaffold, routed work — one wallet.
Lead with outcomes, not commodity line items. Routed calls bill per task; savings show next to each line in-terminal.
Pro
$0.75/ debate
Validate one idea. Get a verdict, sources, and the scaffold to ship V1.
Pro+
Sprint 4$2/ debate
Pro + live evaluation against your shipped metrics.
Studio
Sprint 4$5–19/ sprint window
Recurring sprint orchestration for shipping teams.
Plus ~10% markup on routed calls (gecko_route). You always see the savings vs. premium tier defaults — no surprise line items on the marketing surface.
Cadence
Sprint shipping, not slide-deck quarters.
Each row is a batch of deliverables in the build loop — validate, scaffold, route, reconcile.
- ✓ Now
Now — Sprint 3
Pro debate · scaffold generator · gecko_route (cost-aware routing, live).
- • Next
Next — Sprint 4
Live project reconciliation · sprint orchestration · weekly pulse · Pro+ / Studio tiers.
- ○ Later
Later — Sprint 5+
Feature research agent · MCP marketplace · web dashboard.
Track the repo: github.com/ernanibmurtinho/gecko-claude →
Get waitlist updates
Email for Pro tier on Solana mainnet — install stays the primary path.
Margin note
We publish session economics for builders who want receipts — run gecko-mcp economics <session_id> after any paid session. This is optional depth, not the product pitch.
[ Paid on-chain — Solana mainnet · x402 ]
