convez.ai
MCP ServerIntegração & APIsAvançado
greyseymour

BaseKit — Deploy Onchain na Base via IA

greyseymour/basekit

Deploy e operações onchain na Base direto do prompt da IA

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server com 13 ferramentas + 15 playbooks SKILL.md para deploy e operações onchain na blockchain Base. Permite a Claude automatizar contratos, analisar wallets, simular transações e otimizar gas — tudo a partir de um prompt, sem dependência de terceiros.

Instalação

$npx convez-ai greyseymour/basekit

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Deploy de tokens e contratos inteligentes na Base
  • Análise de wallets e portfólios onchain
  • Simulação de transações e otimização de gas
  • Verificação e debugging de contratos via Basescan
  • Automação de operações DeFi (swaps, multisig, airdrops)

Veredito

Ferramenta especializada e bem-construída para quem trabalha com blockchain Base. A integração MCP é robusta, os playbooks cobrem casos reais (token launch, MEV, Safe multisig), e a arquitetura sem custódia (wallet nunca toca o servidor) é acertada. Recomendado para devs e builders onchain; não serve para quem não trabalha com Web3.

README

A tradução para o português deste README ainda está na fila.

BaseKit

Ship onchain on Base from a single prompt. 15 opinionated SKILL.md playbooks + a 13-tool MCP server, portable across Claude Desktop, Claude Code, Cursor, Codex, Windsurf, and any SKILL.md-compatible runtime.

License: MIT Built on Base MCP

What this is

Two things in one repo:

  1. 15 SKILL.md playbooks — the agent loads them on demand.

    • base-deploy-token, base-token-launch-checklist, base-airdrop
    • base-analyze-wallet, base-portfolio-snapshot, base-onramp-flow
    • base-gas-optimize, base-mev-resistant, base-safe-multisig
    • base-basescan-debug, base-contract-verify, base-event-listener
    • base-frame-build, base-revenue-share, agent-onchain-handshake
  2. An MCP server with 13 tools — the agent calls them directly.

    • Read — get_wallet_profile, get_transaction_trace, portfolio_snapshot, check_token_approvals, gas_now, resolve_basename, check_basename_available, decode_calldata
    • Simulate — estimate_token_deploy, estimate_safe_deploy, simulate_swap, verify_contract
    • Policy — agent_policy_describe (returns the trust manifest)

MIT, free, and built so an agent can install it without a human in the loop.

Install — Claude (one click)

→ Add BaseKit to Claude

Opens Claude.ai's custom-connector modal pre-filled with our hosted endpoint. Works in Claude.ai web, Claude Desktop, iOS, and Android. No clone, no build, no JSON to edit.

  • Endpoint: https://basekit-mcp.workers.dev/mcp
  • Transport: Streamable HTTP (MCP spec)
  • Tools: 13 read-only — your wallet never touches the server
  • Source: worker/src/index.ts — Cloudflare Worker, deploy your own copy in 5 minutes if you'd rather self-host

After connecting, ask Claude:

"What's the gas price on Base right now, and profile vitalik.base.eth"


Install — Claude Desktop (from source, for developers)

If you want full local control or you're contributing to BaseKit, install the stdio server from source. Takes ~3 minutes.

# 1. Clone + build
git clone https://github.com/greyseymour/basekit.git
cd basekit/mcp
npm install
npm run build

Then add to your Claude Desktop config:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "basekit": {
      "command": "node",
      "args": ["/absolute/path/to/basekit/mcp/dist/index.js"],
      "env": {
        "BASESCAN_API_KEY": "your_basescan_key",
        "ALCHEMY_API_KEY": "your_alchemy_key"
      }
    }
  }
}

API keys are free:

Restart Claude Desktop. The 13 tools show up in the hammer menu. Try:

"What's the gas price on Base right now, and profile vitalik.base.eth"

Install — other runtimes (coming with v0.3)

# Claude Code (plugin install — pulls skills + MCP at once)
claude plugin install basekit

# Cursor
cursor mcp add basekit

# Codex CLI
codex mcp add basekit npx -- -y @basekit/mcp

# Windsurf
windsurf mcp add basekit

These land once @basekit/mcp is published to npm.

Local test path (Inspector)

Want to poke the tools directly without a host?

cd basekit/mcp
export BASESCAN_API_KEY="..." ALCHEMY_API_KEY="..."
npx @modelcontextprotocol/inspector node dist/index.js

Opens localhost:5173 — every tool callable from a web UI.

Repo layout

basekit/
├── plugin.json                # Claude plugin manifest
├── README.md
├── LICENSE                    # MIT
│
├── skills/                    # 15 SKILL.md playbooks
├── mcp/                       # @basekit/mcp — 13-tool MCP server
│   ├── package.json
│   ├── src/index.ts
│   ├── dist/index.js          # prebuilt
│   └── tsconfig.json
│
├── brand/                     # palette, type, logo
├── docs/                      # strategy, decisions, permissions
├── launch/                    # tweet thread, cast, listings
├── qa/                        # test fixtures
└── site/public/               # basekit.dev landing

For agentic crawlers

Predictable paths, no human required:

  • /.well-known/agent.json — A2A trust manifest (security disclosure, signing status, delegator contract)
  • /llms.txt — plain-text index of skills + tools
  • /agents.txt — robots-style policy + manifest pointers
  • /skill.md — aggregated playbook (drop into any SKILL.md runtime)
  • /plugin.json — Claude plugin manifest

Philosophy

  • One prompt → one outcome. No "what would you like to do next?" follow-ups.
  • Cost-first. Print gas + ETH cost at current base fee before any tx.
  • Sepolia by default. Mainnet only when the user names the chain explicitly.
  • Never broadcast without confirmation. Side effects require a yes.
  • The agent does the work. Skills are the playbook, MCP is the arm.

Security

This is pre-v1.0 software. Skills are unsigned (signed: false in every frontmatter) until the v0.3 signing infrastructure ships. Read the full threat model at basekit.dev/security before giving an agent broadcast rights. Disclose vulns to security@basekit.dev.

License

MIT. Fork freely. PRs welcome.

Built on Base. Built for agents. Built in the open. 🩵

Ver o original no GitHub

Bom para

Relacionados