Lupin vs Arcade: governed action runtime or shared agent infrastructure

Governance may be the primary product requirement

Arcade’s current documentation describes managed OAuth and tokens, action execution, custom tools and centralized governance. It also documents cloud or self-hosted deployment paths and the ability to govern external MCP servers. Lupin coordinates configured app connections with email, browser, sandbox, phone and proxy resources. Both address agent access, but at different operating boundaries. An enterprise selecting a central action runtime should evaluate policy enforcement and deployment requirements directly rather than treating a broad resource platform as an automatic substitute.

External MCP governance can favor Arcade

A company may already operate several internal MCP servers and want one place to register, filter and audit their tools. Arcade explicitly documents that governance path alongside its own tool catalog. Lupin’s MCP endpoint exposes Lupin’s registered resource operations; its existence does not promise a gateway for governing arbitrary third-party MCP servers. Prefer Arcade when that central runtime requirement is essential. Prefer Lupin when the problem is giving an agent a supported set of resources and app actions under a shared workspace identity.

Account access and action authorization are separate tests

OAuth proves an account connection was authorized for some scopes. It does not prove the user approved the agent’s current proposed change. Arcade’s documentation emphasizes per-action enforcement, so evaluate the exact policy and identity path your application needs. Lupin’s consumer agent records exact app-tool approvals, while external MCP clients must implement their own consent interface. Compare what the operator can inspect: selected account, concrete tool, arguments and resulting effect. A successful login demonstration is insufficient evidence for production action governance.

A reconnect must invalidate stale action context

When a user replaces a personal account with a company account, an old approval should not silently authorize a write under the new identity. Lupin supplies a connection revision and discovered concrete tool version that are checked at execution. Direct runtime users need corresponding account and policy binding in their selected architecture. Include tool definition changes and revoked accounts in the evaluation. Remote descriptions, names and returned app content remain untrusted inputs, even when they arrive through a managed action runtime with a familiar connector name.

Compare governance operations and resource operations separately

Estimate the action runtime, connected identities, policy administration, deployment and audit integration under current Arcade terms. Lupin’s app actions and successful connections are metered alongside separate resource allowances. Arcade may remove substantial governance work for an organization already operating many agents and tools. Lupin may remove resource ownership and coordination work for a product building an agent that needs several infrastructure families. Those benefits are not interchangeable. If self-hosting or an enterprise policy integration is mandatory, make it a hard selection gate and verify the deployment contract before pricing minor API differences.

Build an authorization matrix before a live demonstration

Write three cases: a permitted read, a denied write and an approved action whose account changes before execution. Add an existing internal MCP server if that is part of the requirement. Map the identity and policy decisions to Arcade’s current runtime and Lupin’s exposed controls without using real customer data. Then test with authorized non-production accounts. Prefer Arcade when central tool governance or deployment control is the core need; prefer Lupin when its exact app-action boundary and broader resource model fit the application. Record the denied cases as explicitly as the successful ones.

When switching from Arcade, inventory custom MCP servers, policy integrations, deployment requirements and tool-level authorization rules. Preserve policy fixtures and audit evidence; recreate supported account connections separately. Lupin’s endpoint should not be treated as a transparent replacement for an organization-wide MCP gateway or custom runtime.

Sources & verification

Official documentation and Lupin public documentation/source were reviewed on 2026-11-05. This is an architecture and operating-scope comparison, not a live benchmark, hands-on provider test or independent review of service quality.

• Arcade documentation ↗Checked 2026-11-05
• Lupin connections documentation ↗Checked 2026-11-05
• Lupin bounded agent registration ↗Checked 2026-11-05

Prepared with AI assistance by Lupin editorial. How we research and maintain these guides.