Stellar CLI for Agents
Use the Stellar CLI and CLI Skills to give your AI agents an agent-friendly wallet and native way to inspect, move, and manage assets on Stellar. After following the Quickstart to onboard your agent, you can use natural language to prompt your agent to use the Stellar CLI.
| What You Ask Your Agent | What Your Agent Executes | How It Works Under the Hood |
|---|---|---|
| "Learn how to use the Stellar CLI" | stellar skill | Dumps structured CLI conventions directly into the agent’s prompt/context window. |
| "How much USDC is in alice's account?" | stellar token balance | Reads the balance in smallest units, or as a decimal with --decimal. No key setup needed. |
"Send 20 USDC to G..." | stellar token transfer | Executes an onchain transfer and returns the transaction hash. |
| "Limit agent spending to 50 USDC" | stellar token approve | Grants a capped, time-bound allowance enforced directly onchain by the Stellar network. |
| "Draft a payment for my review" | stellar tx new payment --build-only | Builds an unsubmitted transaction payload to a file for human sign-off or air-gapped execution. |
Agent Ergonomics
Browser wallets are hard for agents. The CLI provides four key agent primitive enhancements:
- Unified Asset Resolution: Interacts with native XLM, SAC (Stellar Asset Contracts), or SEP-41 tokens using consistent identifiers (
CODE:ISSUER, contract ID, or custom alias). - Self-Scaffolding Context: Running
stellar contract initautomatically generates anAGENTS.mdguide in local project directories, giving LLMs instant awareness of local commands and tool rules. - Structured JSON Outputs (
--output json): Dumps typed JSON payloads containing status codes, execution results, and transaction hashes instead of plain text stdout. - Onchain Diagnostic Errors: When a transaction reverts, the CLI returns the contract's diagnostic event trail inside the error envelope. Agents can read why a transaction failed and auto-correct parameters.
Capability Surface
Beyond only using the Stellar CLI as a wallet, your agent can also use the full power of CLI to interact with, build, and deploy contracts. Humans can check out full CLI docs here.
| Functional Area | What Your Agent Can Do | CLI Commands |
|---|---|---|
| System & Agent Context | Check environment health and load usage rules into LLM context windows. | stellar doctorstellar skill |
| Balances & Metadata | Read token holdings, inspect token names, symbols, and decimal precision. | stellar token balancestellar token decimals / name / symbol |
| Transfers & Payments | Move tokens and get back the transaction hash, as a bare line by default or as JSON with --output json. | stellar token transfer |
| Onchain Governance | Set spending limits and audit active delegate approvals. | stellar token approvestellar token allowance |
| Transactions & Contracts | Build raw operations, deploy Soroban contracts, and invoke custom smart contracts. | stellar tx new <OP>stellar contract deploy / invoke |
| Safe Offline Operations | Draft unsigned transactions for human-in-the-loop approval workflows. | stellar tx new <OP> --build-only |
Safety Architecture
Giving an agent financial execution power requires hard guardrails. The CLI relies on protocol-level constraints rather than prompt compliance:
- Protocol Allowance Enforcement: Primary funds remain in your treasury account. Issue your agent a capped allowance using
stellar token approve. If the agent attempts to overspend, the Stellar ledger rejects the transaction. - Human Approval Workflows: Use
--build-onlyflags on sensitive transactions. Your agent formats the payload into a file, leaving final signing authority to a human or air-gapped machine. - Explicit Smallest-Unit Amounts:
token transferandtoken approvetake--amountin the token's smallest unit and do no conversion. Readstellar token decimalsfirst and have the agent show its arithmetic before it sends.
Where to go next
- Skills covers
stellar skill, the built-in guide that teaches your agent the CLI's conventions. - Give your agent a spending limit is the page to read before any real money is involved.
- What bounds your agent is explicit about what the CLI does not protect you from.
- Architecture covers how the read and write paths differ, where state lives, and how to split building, signing, and submitting across machines.
- Output and errors covers which commands emit structured JSON your agent can branch on, because the coverage is not uniform.
- Troubleshooting for when something does not work.