Skip to content

Docs

Claude Cooks is a non-custodial Solana launchpad with an AI agent kitchen attached. Cook a token on a bonding curve, build an agent, bind the two together, and let holders chat with it. This page explains every part of the platform and how it actually works.

Last updated 5 August 2026

What Claude Cooks is

Claude Cooks does two things that meet in the middle. It launches tokens on Meteora's Dynamic Bonding Curve (DBC) on Solana mainnet, and it lets you build AI agents that can be permanently attached to one of those tokens at launch. The token is the access key: holders above a threshold you choose get to talk to the agent.

Nothing is custodial. Your wallet signs every transaction, mint authority and creator fees belong to your keypair, and Claude Cooks never holds funds or private keys.

The sections

  • The Kitchen — the launchpad and directory. Cook a new bonding-curve token, browse everything launched here, and open a launch to trade it and watch its curve fill.
  • Test Kitchen — the agent builder. Define an agent's identity, persona, skills, tool permissions, runtime and holder-access rules before it has a token.
  • The Pantry — the liquidity explorer for Meteora DLMM pools: TVL, volume, fee tiers and liquidity distribution.
  • My Cooks — everything you personally cooked: your agents (including drafts) and your launched tokens.
  • Alerts & Archive — watch conditions on pools you care about, and revisit pools that have gone quiet.

Signing in with a wallet

There are no passwords, emails or social logins. Press Connect Wallet in the header and any installed Solana wallet (Phantom, Solflare, Backpack and other compatible wallets) is prompted immediately; if you have several, pick one from the list.

Sign-in is a signature challenge: the server issues a one-time nonce, your wallet signs that message, and the signature is verified server-side against your public key before a session is created. Signing the challenge costs nothing and never moves funds. Your wallet address is your identity across launches, agents and settings.

Launching a token

A launch is created in The Kitchen. You provide the name, symbol, artwork and supply parameters, and choose a Claude Cooks DBC template that sets the curve shape, fee schedule and graduation threshold. The wizard then builds the Meteora DBC transactions in your browser and hands them to your wallet to sign.

The lifecycle of a launch:

  • Prepping — metadata and artwork are staged; nothing exists on-chain yet.
  • Cooking — the pool and mint are live. Buys move price up the curve, sells move it back down. There is no external liquidity provider; the curve itself is the market.
  • Graduated — the curve filled its quote threshold and migrated to a Meteora DAMM pool, where trading continues as a normal AMM market and LP fees begin accruing.

After the on-chain transaction confirms, the launch is recorded against your wallet so it appears in The Kitchen directory and in My Cooks. Curve progress on a launch page is decoded from live Solana account data rather than a cached summary.

Trading a launch

Each launch page has a native trade panel with Buy and Sell tabs, a slippage control and quotes priced straight from the bonding curve's current state. You approve the swap in your wallet; the transaction is submitted through a server-side RPC proxy and polled until it confirms. Quotes are indicative and can move between quote and confirmation, which is what slippage tolerance is for.

Creator fees, before and after graduation

While a token is still on its curve, trading fees accrue to the creator and can be claimed from the creator fee panel on the launch page.

After graduation the money splits into different pots, and the panel shows each one separately: fees earned by the migrated DAMM LP position, the creator surplus left over from the curve, migration fees, and any leftover tokens still sitting in the DBC vaults. Each claimable pot has its own action and its own wallet signature.

Building an agent

The Test Kitchen builder walks through four things: identity (name, handle, artwork), persona (a system prompt plus a temperament that sets how loose or literal the agent is), skills and tool permissions, and runtime.

Skills are explicit, opt-in and read-only. An agent can be granted things like live token statistics, the top launches in The Kitchen, or the SOL price — it cannot trade, transfer, approve transactions or write to the chain. If a skill is off, the model has no way to call it.

For runtime you either use the hosted model (no key, no infrastructure on your side) or point the agent at your own endpoint, in which case Claude Cooks forwards the conversation to your webhook. The builder can send a test ping to confirm your endpoint answers before you save.

Agents start as drafts. A draft is private to you, editable, and has no token — it stays in My Cooks until you attach one.

Attaching a token to an agent

Attaching happens at launch. When you launch a token from an agent draft, the mint address is bound to that agent permanently: one agent, one token, no reassignment. That permanence is the point — holders can rely on the token being the only key to that specific agent.

Token-gated chat

Each agent has a holder-access rule: the minimum balance of its attached token needed to chat. When someone opens the chat, their wallet's token balance is read on-chain and checked against the threshold, so access follows the token — buy in and you are in, sell and you are out. There is a per-user daily message limit to keep costs and abuse bounded, and conversations are stored per thread so you can return to them.

Where the numbers come from

Pool inventory, TVL, volume and fee tiers come from Meteora's public DLMM data API. Bonding-curve launches are read from Meteora's DBC index and then re-decoded directly from Solana account data. Swap quotes on DLMM pools come from Jupiter's public quote API. Account reads and transaction submission use a Solana mainnet RPC provider, proxied server-side so no provider key ever reaches your browser.

Network stats and pool lists are cached for roughly 30 to 60 seconds; The Kitchen polls every 20 seconds; launch detail pages decode fresh account state on each request. Every screen carries a freshness badge naming its provider and observation time.

We do not fabricate activity, backfill gaps with synthetic points or interpolate missing history. When a provider is unavailable the affected panel says so and stays empty. Atmospheric visuals are presentational only and never encode a number that is not also printed as text. TVL and volume are provider-reported and may lag the chain; prices derived from pool state are spot values that ignore depth.

Common questions

Does Claude Cooks hold my tokens? No. Every action is a transaction you sign; there is no deposit, no escrow and no platform wallet in the path.

Can an agent trade for me? No. Agent skills are read-only by design and there is no code path that lets a model sign or send a transaction.

Can I change which token an agent uses? No. The binding at launch is permanent.

Why is my launch missing from The Kitchen? A launch only appears once its on-chain transaction is confirmed and verified. Until then it is visible to you in My Cooks as an unverified cook.