Quickstart

Use this path when you want one agent action to go through Livy policy before execution.

1. Open the Dashboard

Create or select the project that will own the agent runtime.

You will configure three things:

• a public agent wallet profile
• one guardrail policy
• one runtime API key for MCP

2. Register an Agent Wallet

Go to Wallet Config in the dashboard.

Create a wallet profile with:

• wallet id
• display name
• network, such as devnet or mainnet-beta
• asset or mint
• protected action type, such as solana.native_transfer
• public key

Never paste a private key.

Read Agent Wallet for the matching rules.

3. Create a Guardrail

Go to Guardrails.

Create one guardrail for the action you want the agent to request. At minimum, set:

• the wallet profile it applies to
• the action type
• allowed assets or mints
• per-action value cap
• daily value cap
• status

Read Guardrails before widening limits.

4. Create a Runtime API Key

Go to MCP Setup.

Create an API key for the agent runtime. The secret is shown once, so store it in the server-side environment where the agent runs.

Use these variables:

export LIVY_API_KEY="<runtime-api-key>"
export LIVY_MCP_ENDPOINT="<mcp-endpoint-from-dashboard>"

5. Connect the Agent Through MCP

The protected MCP tool order is:

livy_list_actions
livy_prepare_action
livy_authorize_action
livy_execute_approved_action
livy_get_request_status

Direct wallet sends are outside Livy. They will not create a request, decision, approval nonce, or receipt.

Read MCP for the runtime contract.

6. Add the Livy Skill

If your agent supports skills or instruction files, add the public Livy skill so it uses MCP instead of raw wallet execution.

Read Skills for placement and behavior.

Done State

You are ready for a first protected action when:

• the wallet profile is active
• the guardrail is active
• the runtime key is stored server-side
• the agent calls the Livy MCP endpoint
• the action returns a Livy request id and receipt state