Agent connector terms

Scope and connector definitions

These Agent Connector Terms address automated access through the documented Chase Sets connector surfaces. The native `/mcp` bridge is a separate first-party and operator automation surface where Chase Sets controls the agent host and binds requests to a platform session or operator principal. External agent commerce uses the distinct `/.well-known/ucp` discovery profile and `/ucp/mcp` transport. The Developer Portal documents and links these surfaces; it is not an executable connector and does not imply that every API, tool, or resource is available.

Account attribution and principal questions

Connector activity presented with an authorized credential is evaluated in, and attributable to, the authenticated actor and Account context resolved for that credential. Each call is checked against granted OAuth scopes and permissions, including an Account-scoped actor where the tool requires one. The authorization flow binds the grant to the Account selected by the human authorizing the connector.

The extent to which an account holder is responsible or liable for actions taken by an authorized agent remains unresolved pending qualified counsel review.

Authentication and authorized surfaces

Connector clients use only the documented authentication and authorization surfaces for the connector they call. External UCP clients discover OAuth endpoints from the well-known metadata, use Authorization Code with PKCE, and request the scopes needed for their workflow; the runtime refuses tool calls that exceed the granted permission boundary. Developer Portal documentation does not expand those authorized surfaces or the capabilities currently available.

Permitted purposes and rate limits

Connector requests are subject to the current per-surface Rate-Limit Policy and shipped agent guardrails. The policy can apply named-surface request-volume overrides, and the agent-grant guardrail can refuse writes that arrive too quickly. Numeric limits remain live policy and runtime values rather than fixed promises in these terms.

Connector and market data use

Connector responses expose the catalog and marketplace data returned by tools advertised as available on the current surface. Public discovery tools provide documented read-only marketplace discovery, while the native MCP surface lists only callable capabilities. These product boundaries do not establish terms for downstream retention, redistribution, model training, competitive use, or other reuse.

Agent-assisted transactions

Agent-assisted transactions use the guardrails implemented for the invoked surface. Native tools can require an authenticated actor, permissions, Account scope, exact confirmation, and write idempotency; agent-grant spend checks can apply rail, human-presence, per-order, daily, and monthly boundaries from the resolved mandate. UCP checkout remains a trusted UI handoff unless mandate verification and provider-backed payment conditions are configured, and the shipped checkout and payment path applies only when those gates pass.

Authorization revocation and access questions

The Account consent-management surface lists and revokes Linked Platform Authorizations, and a linked Account can revoke platform consent. Separately, current runtime guardrails refuse operations only when their shipped authorization, confirmation, rate-limit, mandate, or spend checks return a refusal. Those product behaviors do not establish a contractual ground, notice, process, consequence, or Account sanction based on agent conduct.

The grounds, process, and consequences for suspending or revoking agent access, and for suspending, restricting, or closing an Account because of agent conduct, remain unresolved pending qualified counsel review.

Disclaimers and liability questions

The connector surfaces are components of the bounded Chase Sets services described by the applicable public policies and Developer Portal documentation. Any warranties, disclaimers, exclusions, remedy limits, or liability cap for connector use require qualified counsel review; this draft states no numeric cap and does not label a legal allocation as current product behavior.

Changes and versioning

This artifact is version `v1` and remains `counsel-review-required`, with no effective date or counsel approval reference. The canonical policy machinery therefore keeps it non-activatable. Any notice, acceptance, transition, survival, or re-acceptance rule for a later version requires qualified counsel review and is not claimed as current wiring.