# Tesseract > Agent-native wallet setup and execution on Hyperliquid. Bring your own strategy; Tesseract manages scoped wallet access, signing and execution. - [Agent connection guide](/agent-guide.md): Authentication, live permission discovery, execution, recovery and supported market families. - [Deployment health](/api/health): Enabled execution modes. Paper simulates, testnet uses test funds, mainnet uses real funds. ## Connect Use the deployment's app/API origin plus `/mcp`, with Streamable HTTP MCP and `Authorization: Bearer YOUR_AGENT_TOKEN` in secure client configuration. Initialize and call `tools/list` for current schemas. REST uses the same Bearer credential. Owner provisioning/approval happens in the dashboard; the token cannot grant itself authority or withdraw externally. ## Start here 1. Read `get_account_snapshot` for owner, network, saved `agent.policy`, access warnings, balances and unresolved operations. Limits differ by agent; examples aren't permissions or remaining budget. Deployment/exchange constraints also apply. 2. Search `list_markets`; use exact returned identifiers, quote currencies, precision and outcome rules. Follow pagination when needed. 3. For an authorized trade/strategy, use `place_order`. Optional `prepare_order` helps construct price/size strings. `cancel_order`, `set_leverage` and `transfer_collateral` use the same operation pattern and require appropriate authority. ## Execution contract Persist one unique `requestId` and exact payload before each intended action. After timeout or uncertainty, poll `get_operation` or retry the **same ID and payload**; a new ID is a new action. Different parameters conflict. Bindings have no automatic retention expiry; never reuse IDs. A `not_found` lookup doesn't make a fresh ID safe. Read status and actual fill quantities, not just HTTP success. Check observed positions before closing. Tesseract owns signing/nonces. Connection checks are read-only; an authorized strategy doesn't require extra human approval for every trade. Read the guide for deadlines, partial fills, collateral, access expiry and optional preview/execute recovery. Use live schemas and account data for configuration; don't infer capabilities from a fixed tool count or another user's setup.