Connect an agent

MCP (Claude, Cursor, OpenClaw)

Give your own agent Squire's tools over the Model Context Protocol. Your model, your key; Squire only does the on-chain work.

Remote server (recommended)

Squire exposes a Streamable-HTTP MCP server at https://squire-henna.vercel.app/api/mcp, authenticated with your cpk_ key as a bearer.

Claude Code

claude mcp add --transport http squire https://squire-henna.vercel.app/api/mcp -H "Authorization: Bearer cpk_..."

Cursor / any MCP client (mcp.json)

{
  "mcpServers": {
    "squire": { "url": "https://squire-henna.vercel.app/api/mcp", "headers": { "Authorization": "Bearer cpk_..." } }
  }
}

Local stdio package

bash

CLAWPUMP_API_KEY=cpk_... CLAWPUMP_API_URL=https://squire-henna.vercel.app npx clawpump-robinhood-mcp --claude

What the tools cover

Platform: health, stats, the token board, token details, prices, stock tokens, the model catalog.
Agents: create, list, update (model, skills, persona), start/stop, whitelist.
Launch: preview and launch on Pons or long.xyz, sponsored or self-funded; list launches; fee earnings.
Trade & wallet: quote and execute swaps, portfolio, whitelist-gated transfers.
Automations & runs: DCA, limit and sniper strategies; chat with hosted agents; autonomous runs.
Market & skills: buy and sell agents; write and share custom skills.

Money-moving tools are gated

Launch, swap, transfer, claim and buy tools refuse to run until the call includes their confirm_*: true flag. When a tool returns CONFIRMATION_REQUIRED, your agent should tell you what it intends to do and ask, not retry.

Costs

Nothing on Squire's side for thinking: your agent uses your own model and key. Squire only takes its venue fees on launches and trades as usual.