Skip to content

CLAUDE.md

Latest commit

 

History

History
227 lines (178 loc) · 7.98 KB

CLAUDE.md

File metadata and controls

227 lines (178 loc) · 7.98 KB

CLAUDE.md -- AgentLaunch Toolkit

You are working inside the AgentLaunch Toolkit -- a complete system for building, deploying, and tokenizing AI agents on the Fetch.ai ecosystem.

What This Repo Does

This toolkit lets AI agents (including you) do the full lifecycle:

  1. Scaffold agent code from 6 templates
  2. Deploy to Agentverse (Fetch.ai's agent hosting platform)
  3. Tokenize on AgentLaunch (create a tradeable ERC-20 token)
  4. Hand off a link for a human to sign the blockchain transaction
  5. Monitor the token (price, holders, market cap)
  6. Trade via pre-filled links (buy/sell signals)

What's Inside

Package Path Description
SDK packages/sdk/ TypeScript client for every API endpoint
CLI packages/cli/ 10 commands, one-command full lifecycle
MCP Server packages/mcp/ 13+ tools for Claude Code / Cursor
Templates packages/templates/ 6 production-ready agent blueprints

Authentication

Everything uses ONE key: the Agentverse API key from .env.

  • Set it once -- SDK, CLI, MCP, and all examples use it automatically
  • Auth header: X-API-Key: <AGENTVERSE_API_KEY>
  • No wallet signatures needed for API operations
  • Human wallet only needed for on-chain signing (via handoff links)

Environment URLs

The toolkit supports dev and production environments via .env:

Variable Dev (default) Production
AGENT_LAUNCH_API_URL https://launchpad-backend-dev-1056182620041.us-central1.run.app https://agent-launch.ai/api
AGENT_LAUNCH_FRONTEND_URL https://launchpad-frontend-dev-1056182620041.us-central1.run.app https://agent-launch.ai

Set AGENT_LAUNCH_ENV=dev (default) or AGENT_LAUNCH_ENV=production to switch. Or override directly with AGENT_LAUNCH_API_URL and AGENT_LAUNCH_FRONTEND_URL.

Quick Commands

# Build everything
npm run build

# Run tests
npm run test

# Clean all dist/ directories
npm run clean

# Run CLI
npx agentlaunch create                     # Interactive: scaffold -> deploy -> tokenize
npx agentlaunch list                       # List all tokens
npx agentlaunch status 0x...               # Check a token

# Run MCP server (for Claude Code integration)
npx agent-launch-mcp

# Run examples
npx ts-node examples/01-hello-world/deploy.ts

Package Structure

agent-launch-toolkit/
  .env.example              # Copy to .env, add your Agentverse API key
  CLAUDE.md                 # This file (Claude Code reads it first)
  packages/
    sdk/                    # agentlaunch-sdk (TypeScript HTTP client)
    cli/                    # agentlaunch-cli (interactive + scripted commands)
    mcp/                    # agent-launch-mcp (13+ tools for Claude Code)
    templates/              # agentlaunch-templates (6 agent blueprints)
  .claude/
    settings.json           # MCP server config, permissions
    rules/                  # Auto-loaded coding rules
    skills/                 # Slash commands (/build-agent, /deploy, etc.)
  docs/                     # Architecture, references
  examples/                 # Copy-paste workflows

MCP Server (Pre-Configured)

The MCP server is already configured in .claude/settings.json. You have access to these tools:

Tool What It Does
list_tokens Browse tokens (filter by status, chain, category)
get_token Get details for a specific token
get_platform_stats Platform-wide statistics
calculate_buy Preview a buy (tokens received, fee, price impact)
calculate_sell Preview a sell (FET received, fee, price impact)
create_token_record Create a new token -- get handoff link
get_deploy_instructions Get deploy instructions for a pending token
get_trade_link Generate pre-filled buy/sell link
deploy_to_agentverse Deploy Python agent to Agentverse
scaffold_agent Generate agent code from template
create_and_tokenize Full lifecycle in one call
get_comments Read token comments
post_comment Post a comment on a token

Slash Commands

Command Action
/build-agent Scaffold + deploy + tokenize (guided)
/deploy Deploy agent.py to Agentverse
/tokenize Create token for an existing agent
/market Browse tokens and prices
/status Check agent/token status

Architecture

    SDK (TypeScript client)
     |
     +-- CLI (wraps SDK with interactive prompts)
     +-- MCP (wraps SDK as Claude Code tools)
     +-- Templates (agent code generation)
          |
          v
    AgentLaunch API (${AGENT_LAUNCH_API_URL})
     |
     +-- Token operations (create, list, details)
     +-- Market data (prices, bonding curve math)
     +-- Handoff links (deploy, trade)
          |
          v
    Agentverse API (https://agentverse.ai/v1)
     |
     +-- Agent lifecycle (create, upload code, start/stop)
     +-- Secrets management
     +-- Log monitoring

Templates

Template Description Use Case
custom Blank Chat Protocol boilerplate Start from scratch
price-monitor Watches token prices, sends alerts Monitoring service
trading-bot Buy/sell signal generation Trading service
data-analyzer On-chain data analysis Analytics service
research Deep dives and reports Research service
gifter Treasury wallet + rewards Community incentives

Platform Constants (Immutable)

These are baked into the smart contracts. Never change them:

Constant Value
Deploy Fee 120 FET (read dynamically from contract)
Graduation Target 30,000 FET -- auto DEX listing
Trading Fee 2% -- 100% to protocol treasury (no creator fee)
Total Buy Supply 800,000,000 tokens
Default Chain BSC (Testnet: 97, Mainnet: 56)

Fee rule: The 2% trading fee goes 100% to REVENUE_ACCOUNT (protocol treasury). There is NO creator fee split. The contract has no mechanism to send fees to creators.

Agentverse API Gotchas

  • Code upload requires double-encoded JSON: 
    code_array = [{"language": "python", "name": "agent.py", "value": code}]
payload = {"code": json.dumps(code_array)}  # json.dumps required!
  • Use datetime.now() not datetime.utcnow() (deprecated)
  • @chat_proto.on_message(ChatAcknowledgement) handler is required
  • Wait 15-60s after start for agent compilation
  • Agent listing response is { items: [...] } not { agents: [...] }

The Handoff Protocol

Agents NEVER hold private keys. The flow is:

  1. Agent calls API to create token record
  2. API returns handoff link (/deploy/{tokenId})
  3. Agent gives link to human
  4. Human connects wallet, signs transaction
  5. Token is live on-chain

This separation is fundamental to the architecture. Never bypass it.

API Endpoints

Base URL: ${AGENT_LAUNCH_API_URL}
Auth: X-API-Key: <AGENTVERSE_API_KEY>

POST  /agents/tokenize                    Create token -> handoff link
GET   /agents/tokens                      List tokens (paginated)
GET   /agents/token/{address}             Token details
GET   /agents/token/{address}/holders     Holder distribution
GET   /agents/token/{address}/comments    Comments
POST  /agents/token/{address}/comments    Post comment

GET   /tokens/calculate-buy               Preview buy on bonding curve
GET   /tokens/calculate-sell              Preview sell on bonding curve

GET   /platform/stats                     Platform statistics

Agentverse API

Base URL: https://agentverse.ai/v1
Auth: Authorization: bearer <AGENTVERSE_API_KEY>

POST  /hosting/agents                     Create agent
GET   /hosting/agents                     List agents -> { items: [...] }
PUT   /hosting/agents/{addr}/code         Upload code (DOUBLE-ENCODED JSON)
POST  /hosting/agents/{addr}/start        Start agent
POST  /hosting/agents/{addr}/stop         Stop agent
GET   /hosting/agents/{addr}/logs         Get logs
POST  /hosting/secrets                    Set a secret