Clawpump

ClawPump

ClawPump is a Solana platform for AI agents: each agent has a wallet and can trade, run perps, DCA, lend, launch tokens, buy gift cards, send email from its own inbox, and be bought/sold on a marketplace. Its full feature set reaches Hermes through the ClawPump MCP server (134 tools, 10 resources, 10 prompts).

Enabling ClawPump

The tools are not built in — they come from the ClawPump MCP server, which the user installs once:

hermes clawpump setup          # interactive: remote (browser) or stdio (cpk_ key)
# or directly:
hermes mcp install clawpump        # remote OAuth (browser paste cpk_ key)
hermes mcp install clawpump-stdio  # local npx, CLAWPUMP_API_KEY in ~/.hermes/.env

If ClawPump tools are missing, tell the user to run hermes clawpump setup and restart the session — do not try to call the ClawPump HTTP API directly.

Tool naming

Every ClawPump tool is exposed under the MCP namespace mcp_clawpump_<tool> (remote) or mcp_clawpump-stdio_<tool>. Throughout this doc the bare tool name is used (e.g. swap_quote); call the namespaced form your tool list shows.

agent_id

Most tools act on a specific agent. Resolve it once with list_agents, then pass agent_id. If the user set CLAWPUMP_DEFAULT_AGENT (or has exactly one agent), agent_id can usually be omitted. When unsure which agent, ask.

What's available (134 tools, 21 groups)

By default only the 95 read-mostly tools are enabled (including usepod_provision, whose only fund-moving path is double-gated by confirm_deposit). The financial ones below are opt-in via hermes mcp configure clawpump.

⚠️ FINANCIAL & IRREVERSIBLE TOOLS — confirmation is mandatory

These move real funds on-chain or are irreversible. Never call them without the user's explicit, specific go-ahead in this conversation. Quote the exact action (amounts, token, agent) and wait for a clear yes. Several require an explicit confirm flag (e.g. confirmRisk: true, confirm_launch: true, confirm_spend: true, confirm_send: true, confirm_transfer: true, confirm_payment: true) — only set it after the user agrees.

• Trading: swap_execute, dca_create, dca_cancel, limit_order_create, limit_order_cancel
• Perps: perps_account_prepare, perps_trader_register, perps_collateral_deposit, perps_collateral_withdraw, perps_order_execute, perps_order_cancel
• Lending: jup_lend_deposit, jup_lend_withdraw
• Predictions: predictions_open, predictions_close
• Marketplace (funds/ownership): place_bid, accept_marketplace_bid, reject_marketplace_bid, withdraw_marketplace_bid, delist_marketplace_listing
• Token launch: launch_token_gasless, launch_metaplex_genesis_token
• Account / lifecycle: set_external_wallet, delete_agent, delete_automation, delete_custom_skill, cancel_agent_run, remove_integration, remove_from_whitelist
• Agent cards (Laso): agent_card_create (pays USDC from the agent wallet — quote first), agent_card_withdraw, agent_card_cancel, agent_card_reveal (exposes the card number / redemption secret)
• Agent mail: agent_mail_create (one-time ~$2 USDC from the agent wallet), agent_mail_send (outward-facing email; paid per send via x402 from the agent wallet)
• Wallet: wallet_transfer (irreversible on-chain send; destination must be on the agent's whitelist via add_to_whitelist; quote the exact amount, token, and address, then confirm_transfer: true)
• Pay.sh x402: pay_sh_execute_approved (pays USDC from the agent wallet via x402 — always pay_sh_prepare_call first; quote the exact endpoint + price; approval codes expire in 10 minutes; requires the x402 skill on the agent)
• UsePod: usepod_deposit (top up the user's own UsePod pod from the agent's ClawPump wallet, to pay for Hermes/agent inference usage — irreversible on-chain USDC deposit of the full amount, no fee; the user supplies their 16-hex deposit_code from UsePod /v1/register; double-check the code and amount, then confirm_deposit: true)

Always get a quote/preview first (swap_quote before swap_execute, perps_order_preview before perps_order_execute, agent_card_quote before agent_card_create, pay_sh_prepare_call before pay_sh_execute_approved) and show it to the user before executing. Report tx signatures back when a call succeeds.

Resources (read with the MCP resource reader)

Live read-only views under clawpump://:

• clawpump://me — current account/auth status
• clawpump://agents — all your agents
• clawpump://agents/{agentId} — one agent's detail
• clawpump://agents/{agentId}/balance — credit balance
• clawpump://agents/{agentId}/usage — usage & budget
• clawpump://agents/{agentId}/chat — chat history
• clawpump://agents/{agentId}/runs — autonomous runs
• clawpump://marketplace — marketplace listings
• clawpump://models — available LLM models
• clawpump://wallets — wallet summaries

Each resource is also reachable via an equivalent tool (e.g. list_agents, get_balance, get_wallet_summaries), so if the resource reader isn't handy, call the matching tool.

Guided workflows (mirror the MCP prompts)

The server defines these prompt flows; run them with the tools above: get-started, create-agent, setup-trading, launch-token, review-costs, setup-automations, explore-marketplace, setup-dca, explore-predictions, earn-yield.

Common patterns:

• New user: get_account_status → list_agents; if none, create_agent.
• Trade: token_search → swap_quote → (confirm) → swap_execute.
• Launch a token: get_launch_status → confirm details → launch_token_gasless (gasless) or launch_metaplex_genesis_token.
• Check portfolio: get_balance + get_portfolio + get_wallet_summaries.
• Perps: perps_markets/perps_market_data → perps_order_preview → (confirm) → perps_order_execute.
• Gift card: agent_card_search_merchants/agent_card_search_gift_cards → agent_card_quote → (confirm) → agent_card_create → agent_card_status; agent_card_reveal only when the user asks for the card details.
• Agent email: agent_mail_get_address; no inbox → (confirm ~$2 USDC) agent_mail_create; then agent_mail_send (confirm) or agent_mail_list → agent_mail_read.
  • Deliverability (avoid spam): write genuine, complete messages — a specific subject, a greeting, 2–4 sentences of real purpose, and a sign-off. Never send one-word/empty bodies or test-like "hello": low-content mail from a fresh sender domain is the #1 spam trigger. Do not use fake-urgent or deceptive subjects ("IMPORTANT!!", "URGENT") to game filters — that backfires and is deceptive. Tell first-time recipients to mark "Not spam" + add the address to contacts; that trains their inbox fastest.
• Fund an external wallet (e.g. a pay.sh allowance): get_balance → add_to_whitelist (user-approved address) → (confirm) wallet_transfer. See the pay-sh skill.
• Paid x402 API call (agent wallet pays): pay_sh_search → pay_sh_provider_details (price) → pay_sh_prepare_call → (confirm price with user) → pay_sh_execute_approved. The agent needs the x402 skill (update_agent with enabled_skills).
• Discover x402 APIs on Dexter: dexter_search (free, by capability — e.g. "ETH price", "image generation") returns matching resources with their payable resourceUrl, HTTP method, per-call USDC price, and input schema. No separate Dexter wallet.
• Pay ANY x402 URL (open, not just a catalog): x402_pay_check (free — fetch the URL's 402 challenge → price + input schema) → quote the price to the user → x402_pay (pays from the agent's ClawPump wallet, confirm_payment: true, max_amount_usd = the cap the user approved). Works on any x402 endpoint, including a resourceUrl from dexter_search. The wallet signs server-side; spend is hard-capped. Needs the x402 skill (update_agent with enabled_skills).
• Payment funds: check with get_wallet_summaries (usdc_balance).
• Pay your Hermes inference from the ClawPump wallet (UsePod): if the user runs Hermes on their own UsePod pod (USEPOD_API_KEY in ~/.hermes/.env), they can top that pod up straight from their ClawPump wallet instead of funding it manually at usepod.ai. Flow: get_balance (confirm the agent wallet holds enough USDC) → ask for the amount + their 16-hex deposit_code (from UsePod /v1/register) → quote both back → (confirm) usepod_deposit with confirm_deposit: true. The full amount goes on-chain into that pod (no fee); UsePod then draws inference cost from it. This is separate from ClawPump's own hosted chat/credits — it just pays the user's own pod from their wallet.

Setting up Pod (UsePod) — auto flow

When a user picks Pod in the /provider (or /model) picker and it isn't configured, the picker hands you a request to set it up. Pod is a drop-in inference provider whose usage is paid in USDC from the user's ClawPump wallet. Make this as automatic as possible — there is nothing to paste and nothing to restart: the CLI auto-applies the token and switches the session onto Pod the moment usepod_provision returns.

1. Ask the amount + which agent wallet. "How much USDC do you want to put on your Pod? (e.g. 5)" — optionally check get_balance first so you know it can cover it. Then "Which ClawPump agent wallet should fund it?" — call list_agents and let the user pick one; use its agent_id. (If there is only one agent you may skip the question and omit agent_id.)
2. Create + fund in one call — usepod_provision registers a brand-new pod (no signup, no human step) and funds it from the chosen agent wallet. After the user approves the amount, call usepod_provision with that amount, the selected agent_id, and confirm_deposit: true. It returns api_token (the pod credential) + deposit_code and the funding signature.
3. It's applied automatically — in the interactive Hermes CLI, the instant usepod_provision returns, Hermes writes USEPOD_API_KEY (+ USEPOD_DEPOSIT_CODE) to ~/.hermes/.env and switches this session onto Pod. No claw model, no paste, no restart. Watch for Hermes' own confirmation line (✓ Now running on Pod: …) and then tell the user "Your pod is funded and this session is now running on Pod — the next message bills from your pod." Report the funding signature. Do not print the api_token (it's a secret and is already applied). (On non-interactive runtimes — bots/API — the credential is saved but the live switch happens on the next start; say "restart to run on Pod" instead of claiming it switched.)

Notes:

• Top up an existing pod (user already set Pod up before): skip provisioning — ask for the amount + their 16-hex deposit_code, then usepod_deposit with confirm_deposit: true. (usepod_deposit is a fund-mover and is opt-in: if it isn't available, the user enables it via hermes mcp configure clawpump.)
• usepod_provision registers the pod first and funds best-effort, so if funding fails the user still gets a valid api_token (look for funding_error) — the token is still auto-applied; fund it afterward with usepod_deposit.
• This is separate from ClawPump's hosted chat/credits — it pays the user's own pod from their wallet. Always confirm the amount before the on-chain spend and report the tx signature.