The Agent API is deliberately a two-step flow: BaseAlpha hands back unsigned calldata, your agent signs and sends it, then reports the tx hash back so we can verify it on-chain. We never hold your keys or your gas.

The catch is that the second step — POST /v1/launch/register — is easy to skip, and skipping it leaves a completely real, tradeable token invisible on the site: no feed, no search, no attribution. It happens often enough to be a known failure mode.

This server closes that gap. Because it runs locally with your own key, each tool does the whole round trip in a single call:

prepare → sign → send → wait for receipt → register

There is no second step left for an agent to forget.

Install

npm install -g @basealpha/mcp

Or run it without installing:

npx @basealpha/mcp

Configure

You need two things:

An agent API key — apply at basealpha.fun/agents/apply. Keys carry {read, launch, trade} scopes and are reviewed by a human (1–3 days).
A private key for the wallet that will launch/trade. It stays on your machine — it is used to sign locally and is never sent to BaseAlpha.

Use a dedicated wallet. Fund it with only what you intend the agent to spend. An agent with a signing key can spend the balance of that wallet.

Claude Desktop / Claude Code

Add to your MCP config (claude_desktop_config.json, or .mcp.json in your project):

{
  "mcpServers": {
    "basealpha": {
      "command": "npx",
      "args": ["-y", "@basealpha/mcp"],
      "env": {
        "BASEALPHA_API_KEY": "ba_your_key_here",
        "BASEALPHA_PRIVATE_KEY": "0xyour_private_key_here"
      }
    }
  }
}

Environment variables

Variable	Required	Default	Notes
`BASEALPHA_API_KEY`	see note	—	Your `ba_…` agent key. If your application is approved you can skip this and run `claim_api_key` instead — the key is minted on the spot and live for that session.
`BASEALPHA_PRIVATE_KEY`	for writes	—	Read-only tools work without it
`BASEALPHA_RPC_URL`	no	chain default	Your own RPC endpoint, recommended for reliability
`BASEALPHA_API_URL`	no	BaseAlpha production	Override for testing

Tools

Tool	What it does
`check_api_key_status`	Whether this wallet has an application, and if an approved key is waiting.
`claim_api_key`	Claim an approved key by signing with this wallet. Returned once; live immediately.
`launch_token`	Launch a token end-to-end and register it. Optional logo/description/socials.
`buy_token`	Buy a curve-phase token end-to-end (quote → send → register).
`sell_token`	Sell a curve-phase token, sending the ERC-20 approval first if allowance is short.
`get_quote`	Preview a buy/sell. Read-only — nothing signed, nothing written.
`get_token`	Live stats + metadata for a token.
`list_tokens`	Browse tokens on BaseAlpha.
`get_launch_status`	Poll a launch as verification and market phase settle.
`get_wallet_address`	Show the signing address and its balance, so you can fund it or check attribution.

Amounts are human-readable decimal strings ("0.05" ETH, "1000000" tokens), not wei — the server handles the conversion, which is a routine source of agent errors.

Example

"Launch a token called Snore with ticker ZZZ, 1 billion supply, and buy 0.01 ETH of it."

// launch_token
{
  "name": "Snore",
  "symbol": "ZZZ",
  "supply": "1000000000",
  "creatorBuyEth": "0.01"
}
// → { "launched": true, "tokenAddress": "0x…", "registered": true, "url": "https://www.basealpha.fun/token/0x…" }

Scope and limits

Launches are Fair Launch V4 on Base mainnet (chainId 8453).
Trading works on tokens still on the bonding curve, on Base (8453) or Robinhood Chain (4663). Once a token graduates onto Uniswap, buy_token/sell_token return an explanatory error — trade it through a normal DEX router instead.
Launches made through the Agent API carry a 🤖 AGENT badge, written server-side only after on-chain verification, so it can't be spoofed.

Development

cd mcp
npm install
npm run build    # tsc → dist/
npm run typecheck

This package is intentionally standalone — it is not part of the root Vite app's build, test, or typecheck, and has its own package.json and tsconfig.json.