Trencher Agent MCP Server

Trencher Agent is built on top of Charon by @yunus-0x. It extends Charon with a 2-tier multi-LLM cascade screener, trusted KOL signal detection, and a full Next.js web dashboard. See ATTRIBUTION.md for full details.

---

Trencher Agent

AI-powered Solana trench orchestrator. 19 autonomous agents — signal ingestion, LLM screening, Jupiter execution, Telegram control.

⚠️ Codebase is in testing period. Developer does not guarantee any result.

---

Repository Structure

trencher-agent/
├── trencher-web/    # Frontend — Next.js 16 · React 19 · TypeScript · Tailwind CSS v4
├── trencher-core/   # Backend  — Node.js · SQLite · Telegram Bot · Jupiter Ultra
└── signal-server/   # Signal   — Express · Solana WS · Pump.fun · Jupiter Trending

---

trencher-core — Backend Service

The core agent orchestrator. Monitors Pump.fun token flow, enriches candidates, and executes via Jupiter Ultra.

🧠 2-Tier LLM Cascade Architecture

Trencher uses a multi-agent LLM pipeline to balance speed, cost, and analytical depth:

1. Tier 1 (Bulk Screener): Powered by DeepSeek for rapid, high-volume, initial-pass candidate filtering.
2. Tier 2 (KOL Validator & Analyst): Powered by Grok to analyze Twitter/CT narratives, validate trusted KOLs for edge cases, and perform post-trade analysis via /learn.

Stack: Node.js (ESM) · better-sqlite3 · node-telegram-bot-api · @solana/web3.js v1 · ws

Quick Start

cd trencher-core
npm install
cp .env.example .env
# Edit .env with your credentials
npm start

For PM2:

pm2 start index.js --name trencher-agent && pm2 save

Required .env

# Telegram
TELEGRAM_BOT_TOKEN=
TELEGRAM_CHAT_ID=

# Signal server (run signal-server locally or on a VPS)
SIGNAL_SERVER_URL=http://localhost:4000
SIGNAL_SERVER_KEY=your_signal_server_api_key

# Solana RPC
SOLANA_RPC_URL=https://mainnet.helius-rpc.com/?api-key=YOUR_KEY
HELIUS_API_KEY=

# Execution mode: dry_run | confirm | live
TRADING_MODE=dry_run

# Multi-Tier LLM Architecture (Cascade)
ENABLE_LLM=true

# TIER 1: DeepSeek (Bulk Screener)
LLM_T1_BASE_URL=https://api.deepseek.com/v1
LLM_T1_API_KEY=
LLM_T1_MODEL=deepseek-chat

# TIER 2: Grok (KOL & CT Validator, Post-hoc Analysis)
LLM_T2_BASE_URL=https://api.x.ai/v1
LLM_T2_API_KEY=
LLM_T2_MODEL=grok-2-latest

# Live/confirm mode only
SOLANA_PRIVATE_KEY=
JUPITER_API_KEY=

Execution Modes

Strategies

/strategy sniper      # Fee-claim overlap, LLM on, fast exit
/strategy dip_buy     # ATH-distance dip entry, wider TP
/strategy smart_money # Strict holder quality, partial TP
/strategy degen       # Rule-based only, no LLM
/stratset sniper tp_percent 75

Key Telegram Commands

/menu          /strategy      /stratset <id> <key> <value>
/positions     /candidate <mint>
/pnl           /learn <window>   /lessons
/walletadd <label> <address>     /wallets

Storage

All state in trencher-agent.sqlite — positions, decisions, strategies, wallets, lessons. Open positions auto-resume after restart.

---

signal-server — Signal Aggregation Service

Independent microservice that collects token data from Pump.fun (graduated), Jupiter (trending), and Solana WebSocket (fee claims), then serves merged signals via REST API.

Stack: Node.js · Express · @solana/web3.js · ws · axios

Quick Start

cd signal-server
npm install
cp .env.example .env
# Set SOLANA_WSS_URL and API_KEY
npm start

For PM2:

pm2 start src/server.js --name signal-server && pm2 save

See signal-server/README.md for full documentation, API reference, and troubleshooting.

---

trencher-web — Frontend

Landing page and live platform dashboard for Trencher Agent.

Stack: Next.js 16 (App Router) · React 19 · TypeScript · Tailwind CSS v4 · D3.js v7

Quick Start

cd trencher-web
npm install
npm run dev      # http://localhost:3000
npm run build    # production build

Features

• Landing page — pipeline walkthrough, 6 agent layers, 8 feature cards, 3 execution modes, 4 strategy cards, setup docs
• Platform dashboard — live D3 force graph (19 nodes · 44 edges), agent roster with live status, real-time activity log stream, agent detail modals
• D3 animations — constellation (hero + CTA), force-directed agent graph with particle flows
• Visual fidelity — 100% identical to original design; no redesign, no Tailwind rewrites of existing CSS

---

Getting Started

1. Start signal-server first — it collects and serves token signals
2. Start trencher-core — it polls signal-server and executes trades
3. Start trencher-web (optional) — dashboard UI

Make sure SIGNAL_SERVER_KEY in trencher-core/.env matches API_KEY in signal-server/.env.

---

License

MIT