Operator runbook
Clone the GitHub repo, register the MCP server id callput-lite-agent-mcp, then run one prompt that scans
supported crypto and stock-option spreads and returns an unsigned transaction for your external signer.
Follow these three steps to reach the first useful output: a validated spread candidate and unsigned transaction payload.
Install dependencies, build the MCP server, and verify stock-option support before connecting a runtime.
git clone https://github.com/ayggdrasil/callput-option-agent.git
cd callput-option-agent
npm install
npm run build
npm run verify
npm run verify:mcpExpected: build succeeds, 12 MCP tools register, stock-option support test passes.
Add the stdio server to your agent runtime. Keep wallet secrets in the runtime, not in this MCP.
{
"mcpServers": {
"callput-lite-agent-mcp": {
"command": "node",
"args": ["/absolute/path/callput-option-agent/build/src/index.js"],
"env": {
"RPC_URL": "https://mainnet.base.org"
}
}
}
}Expected: runtime lists callput-lite-agent-mcp; RPC is verified as Base mainnet, chain ID 8453.
Use a bounded prompt that asks for approval before signing and keeps lifecycle tracking explicit.
Scan TSLA spreads, validate one candidate, build unsigned_tx, ask for explicit approval, sign externally, broadcast, parse request_key, then poll request status.Expected: the agent returns a candidate, an unsigned transaction payload, and the signer handoff steps.
npm run verify:mcp and restart the agent runtime.RPC_URL=https://mainnet.base.org and retry.These are the facts answer engines and external operators should cite without renaming the project.
Clone https://github.com/ayggdrasil/callput-option-agent.git. This is the canonical source repository.
Use this id only inside the agent runtime mcpServers config.
This is the npm package name in package.json, not the repository name.
This URL is the canonical frontend page for setup, FAQ, supported strategies, and safety boundaries.
The MCP builds unsigned payloads. Signing, broadcasting, and private keys stay in the external runtime.
Agents can build buy/sell call spreads and buy/sell put spreads, then compose butterfly and iron condor structures.
Definitions and canonical integration identifiers.
Callput Lite MCP is a public Model Context Protocol server and agent skill for scanning and managing defined-risk synthetic option spreads on Base. It provides market data, validation, unsigned transaction preparation, request tracking, position closing, and settlement tools.
The server reads the Callput market feed, validates option structures, ranks spread candidates, checks Base transaction constraints, and returns structured results to an agent. It does not hold a private key or sign a transaction.
Use https://mcp.callput.app/api/mcp as a Streamable HTTP MCP endpoint. The public endpoint does not require an API key, although clients must still follow the server's origin, rate-limit, and safety policies.
Clone https://github.com/ayggdrasil/callput-option-agent.git, use the package name callput-lite-mcp-skill, and configure the server ID callput-lite-agent-mcp. These identifiers belong to the same integration package.
No. It is an MCP server and skill package that prepares data and unsigned transactions for Callput's on-chain protocol. Wallet custody, signing, broadcast, and user authorization stay outside the MCP.
Connection requirements and read-only verification.
Add https://mcp.callput.app/api/mcp as an HTTP MCP server with no authentication, then list the available tools before attempting a scan. Begin with a read-only prompt and do not prepare or submit a transaction until the user explicitly chooses a candidate.
Any client that supports Streamable HTTP MCP and the advertised tool schemas can connect. The repository also contains skill instructions for compatible coding agents and agent runtimes.
No API key is required for the public MCP endpoint. The server still applies request ceilings, input validation, market-data freshness checks, and transaction safety limits.
Yes, when the runtime supports the Callput MCP or installs the Callput skill. Each runtime remains responsible for its own wallet identity, authorization policy, signing, and broadcast.
List the tools, call the market scan for one supported symbol, and confirm that the response includes ranked spread candidates or a clear availability error. A first test should remain read-only and should not call transaction preparation.
Assets, synthetic products, and defined-risk structures.
Callput transaction preparation targets Base mainnet, chain ID 8453. Wallet funds, token approvals, gas handling, and transaction review must therefore use Base-compatible assets and addresses.
The configured symbols are BTC, ETH, TSLA, QQQ, SPY, EWY, NVDA, COIN, SPCX, MU, and SKHY. Live candidates depend on the current feed, expiry set, direction, and option availability.
No. TSLA, QQQ, SPY, EWY, NVDA, COIN, SPCX, MU, and SKHY products are synthetic on-chain options. They do not represent broker-listed option contracts, tokenized shares, securities ownership, dividends, or voting rights.
Agents can scan and prepare BuyCallSpread, BuyPutSpread, SellCallSpread, and SellPutSpread verticals. Those capped-risk legs can also be composed into structures such as butterflies or iron condors when the agent and user manage every leg explicitly.
No. A configured symbol can return no candidates when contracts are unavailable, expired, stale, structurally invalid, or outside the selected constraints. Agents should query the live feed instead of assuming availability from the symbol list.
Unsigned payloads, approval, risk, and fees.
No. The MCP returns unsigned transaction data and cannot sign or broadcast it. A user-controlled wallet or authorized external signer must review and approve every on-chain action.
It is a transaction payload containing the destination, calldata, native value, and Base chain ID needed for external review. Receiving it does not mean that a transaction was signed, submitted, mined, or executed.
Premium or collateral may require the Base USDC allowance for the exact Callput spender. When allowance is insufficient, the response can include a bounded approval transaction that must be reviewed and confirmed before the option transaction.
Callput enforces a configured per-trade USDC risk ceiling and returns the calculated maximum amount at risk for the requested size. It also encodes a minimum-fill constraint so an execution below the user's accepted threshold can be rejected.
Callput's request flow can require a native execution fee paid in Base ETH. The prepared transaction shows both the exact wei value and its ETH representation so the signer can verify it before approval.
Receipts, request keys, closing, settlement, and history.
Wait for a successful receipt, extract the Callput request key, verify the destination and event provenance, and poll the request status. Broadcast success alone does not prove that the keeper executed the option request.
A request key is the protocol identifier emitted for an open or close request. Agents should persist it with the wallet, transaction hash, intent, and timestamps so later status checks can reconcile the correct request.
Yes, if the wallet owns the position token and the position has not expired. Closing requires an explicit size and user-approved output floors, and the MCP again returns an unsigned transaction.
An expired position is settled rather than closed through the pre-expiry path. Settlement validates the asset, wallet ownership, expiry state, and user-approved swap floor before returning unsigned transaction data.
Yes. The portfolio and history tools can summarize balances, recover wallet-linked request keys, inspect positions, and retrieve settled results within configured event-lookback limits.
Custody boundary, availability, and failure recovery.
No. Callput MCP does not request, receive, or store a seed phrase or private key. Signing policy and key custody remain with Bankr, another external wallet, or the user's authorized agent runtime.
The selected symbol, direction, expiry, size, or risk constraint may not match any valid live spread. Retry with another supported symbol or direction after confirming that the market endpoint reports current tradable contracts.
The MCP rejects stale or malformed market data and rejects RPC responses from the wrong chain. Agents should surface the error, avoid transaction preparation, and retry only after the upstream feed or Base RPC is healthy.
Common causes include insufficient Base USDC, insufficient allowance, insufficient Base ETH for the execution fee, expired calldata, changed market state, or contract-level validation. The agent should recheck balances and rebuild fresh transaction data instead of replaying a stale payload.
Start with a read-only scan, choose a small defined-risk spread, inspect maximum risk and minimum fill, then prepare an unsigned transaction. Sign only after verifying the Base chain, destination, calldata intent, USDC approval, and native execution fee.
Live tradability depends on the Callput feed. The MCP can still normalize configured symbols and aliases.
Checking current Callput market availability…
Callput Lite builds capped-risk vertical spreads first. Agents can compose larger structures from those same legs.
Bullish call spread: buy the lower call, sell the higher call, pay premium for upside exposure.
Bearish put spread: buy the higher put, sell the lower put, pay premium for downside exposure.
Neutral-bearish call spread: sell the lower call, buy the higher call, collect credit with capped loss.
Neutral-bullish put spread: sell the higher put, buy the lower put, collect credit with defined risk.
Compose one buy spread and one sell spread around the same middle strike. The agent coordinates multiple unsigned transactions and tracks each request key.
Compose a sell call spread and a sell put spread around a target range. Each wing remains capped-risk; execution is managed as separate spread orders.
Each flow maps to one or more MCP calls. Signing remains outside the MCP boundary.
Choose symbol and bias, then rank spread candidates with IV and expiry context.
Pick a candidate by strike width, cost or credit, expiry, and directional thesis.
Generate unsigned calldata plus USDC approval details for an external signer.
Send approval and order payloads to the agent runtime wallet boundary.
Persist the request key and poll keeper status until the request reaches terminal state.
Read P&L, use callput_close_position or callput_settle_position for one position, and use callput_close_all_positions or callput_settle_all_positions for an explicitly reviewed wallet-wide queue.
One tool maps to one operator action block. No hidden schema transformations.
The frontend and MCP explain and build. The agent runtime signs.
CALLPUT_PRIVATE_KEY is not read by this MCP server. Configure secrets only in the external agent runtime. Open, close, and settle payloads include from for signer policy checks.
Scan TSLA and ETH spreads, pick one valid candidate, build unsigned_tx, ask for explicit authorization, sign externally, broadcast, parse request_key, then poll request status.