AgentPad

Registre officiel IA et agents

AI-agent token launchpad on Base: quotes, tokens and unsigned launch/buy/sell/claim/vote txs.

Installation

Ajoutez ceci à la configuration MCP de votre client (Claude Desktop : claude_desktop_config.json ; Cursor : .cursor/mcp.json ; VS Code et les autres utilisent la même forme « mcpServers »).

Remote (streamable-http)
Claude Code
claude mcp add --transport http agentpad 'https://api.agentpadbase.com/mcp'
Configuration JSON (Claude Desktop, Cursor, VS Code)
{
    "mcpServers": {
        "agentpad": {
            "type": "http",
            "url": "https://api.agentpadbase.com/mcp"
        }
    }
}

Documentation

Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.

Sommaire
  1. AgentPad MCP
  2. Humans: tell your agent to launch it
  3. Use it
  4. Tools
  5. How AgentPad works (short)
  6. Live numbers (read from Base mainnet, 2026-10-25 12:00 UTC)
  7. Links
  8. License

AgentPad MCP

Agents launch, anyone trades. AgentPad is a token launchpad on Base where AI agents launch the coins. This repo is its MCP server: it lets any MCP client (Claude, ChatGPT, Cursor and others) read AgentPad and build transactions for it.

Open trading is live: any wallet can now buy and sell on AgentPad v2 curves. Launching is still agent-only.

Bonding-curve tokens are tiny, illiquid and can go to zero. Nothing here promises a price, income or an airdrop.

No keys, ever. Every tool is either a read or returns an unsigned transaction {to, data, value}. The server never asks for, stores or sees a private key; you sign with your own wallet.

Humans: tell your agent to launch it

Launches are agent-only, but you don't need to write code. Add this MCP server to the AI agent you already use (ChatGPT, Claude, Cursor...) and tell it something like:

"Use the AgentPad MCP to launch a token called X with ticker Y on Base, then give me the unsigned transaction to sign."

Your agent registers an agent identity, builds the launch transaction, and you sign it in your own wallet. Guide: https://agentpadbase.com/tell-your-agent.md

Use it

Remote (no install). Streamable HTTP, no auth, no keys:

https://api.agentpadbase.com/mcp
{ "mcpServers": { "agentpad": { "type": "streamable-http", "url": "https://api.agentpadbase.com/mcp" } } }

Local (stdio, Node 18+):

git clone https://github.com/moneymanai/agentpad-mcp && cd agentpad-mcp && npm install && node server.mjs
{ "mcpServers": { "agentpad": { "command": "node", "args": ["/path/to/agentpad-mcp/server.mjs"] } } }

Docker (stdio): docker build -t agentpad-mcp . && docker run -i --rm agentpad-mcp (no secrets; AGENTPAD_API defaults to https://agentpad.thevibehosting.com).

Optional env: BASE_RPC_URL (default https://mainnet.base.org), AGENTPAD_REFERRER (an address that gets the one-level referral fee). No API key needed.

Tools

GroupTools
Readagentpad_info, list_tokens, get_token, get_roadmap, agent_status, quote_buy, quote_sell, list_open_votes, check_vote_eligibility, honest_volume, token_transparency
One callagentpad_buy / onboard_and_buy (on v2 curves any wallet buys in one tx; v1 curves add agent identity -> pass first; also as an EIP-5792 wallet_sendCalls batch)
Unsigned txsbuild_register_erc8004_tx, build_launch_tx, build_team_launch_tx, build_buy_tx, build_sell_tx, build_sell_with_permit_tx, build_claim_tx, build_vote_tx, build_claim_milestone_tx, build_finalize_vote_tx, build_release_milestone_tx, build_burn_expired_tx

2.8.0: agentpad_launch routes through the AgentPad launch endpoint; same inputs and output as 2.7.0 (add-only).

2.7.0: pending-launch tools (inactive until pad v2.4), add-only.

2.6.0: one-tool launch and holder checks. agentpad_launch {wallet, name, symbol, imageUrl | imageId | imageBase64 | imagePath} returns every unsigned call your wallet still needs, in order: ERC-8004 identity (if missing), then AgentPad pass (if missing), then launch. It also returns an EIP-5792 batch and a cast script. The launch fee is 0, and gas for all of it is about 0.00002 ETH. Every error says the next step. holder_check {token, wallet, min} tells a seller whether a wallet holds at least min tokens, so token holders can get free or cheaper calls (guide: https://agentpadbase.com/llms.txt#free-if-you-hold). Add-only: no existing tool changed.

2.5.0: sell approve step, ready for AgentPad pad v2.3. From v2.3 on, a curve sell moves your tokens with the standard ERC-20 allowance, so it needs an exact-amount approve first. build_sell_tx takes an optional wallet; it reads the live pad implementation and, when the pad sells with the allowance and the wallet's allowance is short, returns approveTx (token.approve(pad, exact amount)) and steps (approve, then sell). Send approveTx, wait for it to confirm, then send tx. tx is still the sell, and on the current pad nothing changes (approveTx: null). For tokens launched from v2.3 on (EIP-2612), approval.permit.typedData lets you sign once and call the new build_sell_with_permit_tx for a one-transaction sell. Approve only the exact amount, never unlimited.

2.4.2: wording cleanup in tool descriptions, risk text and docs. No tools, inputs or outputs changed.

2.4.1: official site links (docs, default agentURI, help text) now point to https://agentpadbase.com; the API and remote MCP stay at https://agentpad.thevibehosting.com.

2.4.0: every launch now needs a token picture. build_launch_tx and build_team_launch_tx take imageUrl, imageBase64, imageId or (stdio) imagePath; without one they return IMAGE_REQUIRED, and a bad file returns IMAGE_INVALID (PNG, JPEG, WebP or GIF only, at least 64x64, max 2 MB, checked by magic bytes and a full decode; SVG/HTML rejected). metadataURI is now optional because AgentPad hosts metadata with your picture. New tools: upload_token_image (validate and re-host as 512x512 PNG, returns imageId/imageUrl plus a computed, unpinned IPFS CID), get_token_image (current picture or generated fallback, history, metadata JSON) and update_token_image (the creator changes the picture at any time, also after graduation, with an EIP-191 signature over a nonce + timestamp message; old versions are kept; the tool never touches keys). Free, no paid pinning.

2.3.0: build_buy_tx takes an optional wallet. With it, the result adds preflight (agent pass, ETH balance vs value + gas, eth_call simulation, willSucceed, problems, next), so an agent knows before signing whether the buy will go through. Without it, a hint explains the pass rule and the one-call path. Output is otherwise unchanged.

How AgentPad works (short)

  • Agents launch. Only wallets with an on-chain agent identity (ERC-8004, or an AgentPad pass) can launch. Launch fee 0, Base gas only.
  • Anyone trades. Since 2026-10-26 UTC, any wallet can buy and sell on AgentPad v2 curves, with no agent pass needed to buy. Launching stays agent-only. v1 curves (such as $APAD) still need a pass to buy.
  • Fees. 1% per curve trade; the creator team earns 0.50% of every buy and sell.
  • Graduation. At about 1 ETH of net buys the curve graduates to Uniswap V2 and the LP is locked in a contract with no withdraw function.
  • Roadmap-locked creator allocation. The creator's 5% unlocks only on public milestones (graduation, price TWAPs, one-agent-one-vote polls); a missed deadline burns that tranche.

Live numbers (read from Base mainnet, 2026-10-25 12:00 UTC)

19 registered agents, 3 tokens (including PULSE, the first token launched by an outside agent), 19 curve trades, about 1.148 ETH of curve volume, no graduations yet. Early and small. Current figures: https://agentpadbase.com/stats.json

License

MIT. See LICENSE and SECURITY.md.

Questions fréquentes

Qu'est-ce que AgentPad ?

AI-agent token launchpad on Base: quotes, tokens and unsigned launch/buy/sell/claim/vote txs.

Comment installer AgentPad ?

Il fonctionne comme serveur distant. Dans Claude Code : claude mcp add --transport http agentpad 'https://api.agentpadbase.com/mcp'. Pour les autres clients, collez la configuration JSON de cette page dans leurs réglages MCP.

Faut-il une clé API pour AgentPad ?

Son entrée au registre n'indique aucune clé ni secret obligatoire. Certaines fonctions peuvent tout de même demander un compte auprès du service auquel il se connecte.

AgentPad est-il open source ?

Oui. Son code est sur GitHub sous licence MIT.

Qui maintient AgentPad ?

Il est publié par moneymanai. AI.tn le référence sans le maintenir ni le recommander.

Issu du registre MCP officiel. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.