STRATEGY LAYER · DEFI · MAINNET
Your agent shouldn't trade in the dark.
gecko_trade_research turns one question — "Should I deposit USDC into Kamino?" — into a cited verdict with surviving dissent. Settled per-call in USDC on Solana via x402.
▸ A strategy you can buy, audit, or stake on. No subscription.
curl -fsSL https://app.geckovision.tech/install.sh | bashRead https://app.geckovision.tech/skill.md and follow the instructions.Verify the strategy layer is live without installing anything:
curl -fsSL https://app.geckovision.tech/test.sh | bashClaude Code · MCP · Solana mainnet · ~30s install
SECTION 01
How it works
Drop a .mcp.json in your project.
One file, one endpoint:
https://api.geckovision.tech/mcp/. No SDK, no auth detour.Open Claude Code (or any MCP-aware agent).
Auto-discovers the server. 19 Gecko tools mount, including
gecko_trade_research.Ask a real question.
Use gecko_trade_research on Kamino with the question: should I deposit USDC?Pay per call — $0.25 basic, $0.75 pro.
x402 challenge → USDC on Solana → settled in ~1.6s. The pro tier adds a CoinGecko-OHLCV
backteston the strategist's intent.Get a verdict you can act on.
{ "verdict": "pass", "confidence": 0.75, "surviving_dissent": [...], "citations": [...] }— seven specialists debate, the dissenters survive, the receipt is on-chain.
SECTION 02
What you actually pay
$0.25 basic, $0.75 pro. Per call. Real costs broken down. No subscription, no seat fees.
| Line | Real cost |
|---|---|
| 7-agent panel (gpt-4o-mini, AG2 GroupChat) | $0.05 |
| Voyage embeddings (1024-dim, MongoDB Atlas) | $0.008 |
| Hybrid retrieval (vector + lexical) | $0.003 |
| Pro: CoinGecko OHLCV backtest | $0.010 |
| Total COGS (pro) | $0.071 |
| Pro retail price | $0.75 |
| Margin | 90% |
Every paid call returns a tx_hash you can verify on Solscan. The settlement is the receipt.
On-chain receipt: every paid call is a real Solana tx signature. Your agent's strategy is auditable by another agent.
SECTION 03
Seven specialists. One verdict. Surviving dissent.
Every call to gecko_trade_research runs an AG2 GroupChat panel. Each voice debates from a domain perspective. The dissenters survive into the response — so the agent calling Gecko sees the disagreement, not the consensus laundering.
technical_analyst
Reads TVL, utilization, yield curves, oracle staleness. The data perspective.
fundamental_analyst
Audits, reserve mechanics, governance, integration partners. The protocol perspective.
sentiment_analyst
Twit.sh, paysh news catalog, recent governance chatter. The market mood.
risk_manager
Stress tests the trade. Surfaces tail risk. Most likely voice to dissent.
strategist
Forms the entry/exit/horizon intent. Pro tier hands this to a CoinGecko OHLCV backtest.
bull_bear_debater
Adversarial closing round. Forces the strongest counter-argument before the verdict.
coordinator
Synthesizes seven voices into one verdict + confidence + surviving dissent.
SECTION 04
Roadmap
- ✅ shippedv1Mainnet x402 settlement. 7-agent panel. Citations[] first-class on the wire. Pro tier with backtest. Hybrid retrieval over Mongo Atlas Voyage embeddings.
- ⏳ in flightv2Listing on the pay.sh catalog. Builder integrations from the Solana hackathon. ALB timeout hardening. Upto-scheme x402 buyer for LLM-billed providers.
- 🔜 nextv3Multi-vertical strategy layer beyond DeFi. Agent-to-agent verdict chaining. Bench leaderboard for paid agent skills, scored by surviving dissent. Lower price floors via cached retrieval.