VeTrade VeTrade AI integration docs for VeChain trading AI / Agents
Machine-native surface

Changelog

Version history for the public AI docs, MCP server, and machine-native VeTrade surfaces.

Remote MCP server MCP manifest OpenAPI artifacts

2026-04-17

  • changed the Anthropic managed-agent bootstrap to preload the helper and runtime archive as mounted session files, so first-run live execution no longer depends on a session-side download
  • added empty MCP resources/list and resources/templates/list responses for generic client compatibility, while keeping the hosted VeTrade surface tools-first
  • added instructive resources/read not-found behavior instead of falling through to a generic unknown-method failure
  • remapped JSON-RPC -32601 probe noise to unsupported_method in admin telemetry so compatibility misses stop inflating server-error reporting
  • bumped the published MCP metadata version to 0.5.2

2026-04-16

  • added optional local VIP-191 fee delegation to localExecution.mjs and the OpenAI SDK live execution tools, with session default defaultGasPaymentMode and per-call gasPaymentMode
  • kept sponsor key material local-only through GAS_PAYER_PRIVATE_KEY, GAS_PAYER_PRIVATE_KEY_FILE, or gasPayerPrivateKeyFile
  • updated preview and submit output to report fee-payer metadata and name the correct wallet when delegated VTHO is missing
  • taught the managed-agent execution bootstrap to upload and mount an optional gas-payer key through the raw Anthropic Files API path, and taught the helper to prefer explicit signer or sponsor key files over env keys
  • taught the managed helper swap path to accept amount as a compatibility alias for amountIn, and trimmed the OpenAI live swap example down to a compact operator-facing result
  • refreshed the AI docs, managed-agent examples, and published public mirrors for fee delegation and managed execution behavior

2026-04-15

  • added public hosted transfer, swap_transfer_exact_in, and swap_transfer_exact_out tools plus POST /v2/transfer-plans
  • added recipient-only .vet resolution through the official vet.domains Resolve Utils contracts and attached canonical recipient resolution metadata to the returned execution packages
  • extended the local execution helper with transfer, swap-transfer-exact-in, and swap-transfer-exact-out one-shot commands for Anthropic and Codex managed flows
  • extended the OpenAI SDK with execute_transfer, execute_swap_transfer_exact_in, and execute_swap_transfer_exact_out, plus recipient_optional policy support for recipient-bearing transfer actions
  • filtered overlapping hosted task tools out of live-capable OpenAI agents so local execute_* tools own the signing path when enabled
  • surfaced field-level MCP validation issues through the SDK client and JSON-RPC error message text
  • fixed compact settlement parsing to preserve the final recipient and keep intermediate swap-event recipients in reportedRecipient
  • taught the local swap-transfer helper to accept amountIn and amountOut as compatibility aliases while keeping amount as the canonical field
  • taught the local swap-transfer helper to accept inputSymbol and outputSymbol as compatibility aliases while keeping inputToken and outputToken as the canonical fields
  • removed duplicate field-level validation text from the OpenAI SDK MCP client when the hosted JSON-RPC message already includes the issue summary
  • bumped the MCP metadata version to 0.5.1 and refreshed the AI docs to cover recipient-aware transfer planning and local/live tool boundaries

2026-04-14

  • added one-shot quote parity with local quote_swap in the OpenAI SDK and helper quote-swap in Anthropic/Codex managed flows
  • fixed the OpenAI SDK signer-bound limit-order task flow to use the canonical hosted address field, which restores one-shot create and cancel parity with Anthropic managed agents
  • corrected the Anthropic AI and MCP docs to reflect the published always_allow tool permissions instead of the older always_ask flow
  • documented the local execution state cache so repeated identical live demos can use a fresh CODEX_HOME or session when a new submission is required

2026-04-13

  • promoted swap, create_limit_order, and cancel_limit_order to the primary public MCP entry points for instant-on VeTrade planning and preflight
  • added canonical execution-package responses with normalized request data, resolved tokens, bundle, validation, readiness, economics, continuity, warnings, and local execution handoff instructions
  • made address the canonical required signer-bound field for the hosted swap, create_limit_order, and cancel_limit_order tools while keeping recipient and maker as compatibility aliases behind the endpoint
  • removed the uneconomic-execution override from the canonical execution handoff and local execution policy path so gas-dominates-trade results stay advisory instead of blocking automation
  • taught the public MCP token-selector validators to accept singular selectors alongside arrays and to treat empty selector arrays as absent instead of a validation trap
  • updated the local execution helper and OpenAI SDK integration to consume the hosted execution package directly, while deriving the signer address locally from PRIVATE_KEY or privateKeyFile
  • refreshed the MCP metadata and AI docs around the instant-on hosted flow, the local-signing boundary, and host-side schema troubleshooting

2026-04-12

  • introduced agents/openai-sdk/ as a self-hosted Node reference integration for the official OpenAI Agents SDK
  • fixed the OpenAI hosted MCP tool label in agents/openai-sdk/ so the Responses API accepts the published VeTrade agent configuration instead of rejecting server_label values with spaces
  • marked the OpenAI local execution tools in agents/openai-sdk/ as non-strict so optional inputs such as slippageBps, maxHops, and outputMode stay compatible with the current Responses API function-schema rules
  • added a typed execution policy for OpenAI local execution tools covering allowed actions, allowed pairs, max input, slippage, freshness, and uneconomic-execution overrides
  • kept the public MCP surface read-only and reused the published localExecution.mjs helper as the canonical local signing and submission path
  • introduced summary-first MCP inspect tools: inspect_swap_readiness, inspect_limit_order_creation_readiness, and inspect_limit_order_cancellation_readiness
  • published the OpenAI SDK examples and docs alongside the existing Codex and Anthropic paths
  • bumped the MCP metadata version to reflect the new additive inspection tools and OpenAI SDK documentation

2026-04-10

  • fixed the published localExecution.mjs runtime-loader error so incomplete runtime bundles now fail with instructive compatibility details instead of root is not defined
  • fully versioned the vendored execution runtime snapshot and added runtime archive integrity checks so the public runtime.tar.gz cannot ship without the required VeChain and ethers files
  • added resolve_supported_tokens so MCP clients can resolve token symbols or addresses into a token metadata list without scanning the whole registry first
  • taught localExecution.mjs to read signer material from privateKeyFile, and added an economics guardrail so tiny VTHO-denominated smoke tests require an explicit override when gas dominates the trade
  • added a dedicated signer command to localExecution.mjs so managed agents can resolve the signer address through the helper instead of improvising Python, Node, or shell crypto snippets
  • added a high-level swap command to localExecution.mjs so agents can resolve tokens, prepare the bundle, validate it, and submit a symbol-driven live swap in one serial helper call
  • added submit-atomic as the canonical managed-runtime helper command so agents can pipe one execution payload directly into the helper instead of splitting write and submit steps
  • tightened the managed-agent execution guidance to keep live swaps serial, keep narration minimal, and pass /mnt/session/uploads/vetrade/private_key.txt through privateKeyFile without ever printing the mounted signer contents

2026-04-09

  • introduced list_supported_token_refs as a compact supported-token discovery tool for MCP clients
  • introduced get_token_details so clients can fetch richer token metadata only for the token they actually need
  • documented the refs-first token workflow and Anthropic Managed Agent MCP tool-confirmation behavior
  • packaged scripts/localExecution.mjs with the VeTrade skill so Claude and Codex can preview, execute, and re-check local transactions through one helper
  • added managed-agent bootstrap scripts that create an Anthropic custom skill for the VeTrade execution helper and start sessions with the signer file mounted automatically
  • added a one-shot submit command to scripts/localExecution.mjs, auto-relaunch through --use-env-proxy when proxy env vars are present, and docs that steer new agents away from redundant quote_swap calls for live execution
  • updated the execution helper to prefer signer-facing settlement data from receipt transfers when swap events point at intermediate settlement addresses
  • clarified the managed-agent execution path to write payload JSON to a file before calling localExecution.mjs submit, exposed the mounted signer file id in the session bootstrap script, and documented that tiny swaps can succeed technically while still being economically poor because gas dominated the trade
  • taught the execution helper to recover swap intent from quote metadata when managed runtimes trim bundle.decodedCalls, which restores input-token balance checks for managed-agent submissions
  • bundled a pinned execution runtime next to localExecution.mjs, removed the managed-agent npm install bootstrap path, and published runtime timing metadata plus runtime.tar.gz for cold-start installs
  • switched the supported Anthropic managed-agent execution path to a plain MCP agent plus session-side helper/runtime download, because the packaged runtime currently exceeds Anthropic's custom-skill upload limit
  • refreshed the bundled runtime to a helper-compatible VeChain SDK version, added a fee-history fallback for stale runtimes, and surfaced instructive compatibility errors instead of raw method-missing crashes

2026-04-01

  • made the VeTrade Codex skill execution-capable for local swaps and local limit-order create or cancel flows with PRIVATE_KEY
  • kept the public MCP server and HTTP APIs quote, prepare, validate, and read only
  • updated the AI docs and examples to distinguish local skill execution from the prepare-only server boundary

2026-03-25

  • introduced POST /v2/limit-order-plans
  • introduced GET /v2/limit-order-plans usage guidance for machine clients
  • introduced GET /v2/limit-orders and GET /v2/limit-orders/{orderId}
  • introduced limit-order MCP tools, schemas, OpenAPI, and examples
  • broadened the VeTrade AI surface from swap-only to swap plus limit-order trading

2026-03-24

  • introduced POST /v2/swap-plans
  • introduced GET /v2/swap-plans usage guidance for machine clients
  • introduced public JSON Schemas and OpenAPI
  • introduced remote MCP tool surface
  • made https://mcp.vetrade.vet the canonical MCP endpoint and kept /mcp as a compatibility alias
  • introduced Codex skill guidance
  • removed public auth requirements for the current beta and added Codex-native skill metadata

Compatibility

  • /v1 remains available for legacy consumers
  • /v2 is the machine-native contract for new agent integrations