Authorize agent tools by the exact effect they can cause
Distinguish ordinary lookups, sensitive writes and provider actions that need explicit human review.
Lupin field guides
Agents & MCP
Agent registration, workspace ownership and tool protocols.
20 published articles
Build a framework-independent Lupin HTTP client
Use standard HTTPS, explicit error handling and durable operation records without depending on a particular agent framework.
Check provider readiness separately from permissions and quotas
Discover configured capabilities, then verify that the current workspace may actually use the requested resource.
Choose projects and inbox scopes for agent tenancy
Map customers, workflows and shared resources to the authorization boundaries the API actually enforces.
Claim an agent workspace through a verified human session
Associate a human owner with an autonomous workspace without exposing recovery credentials or upgrading the original key.
Connect to Lupin MCP with initialize, discovery and a read
Negotiate the implemented protocol, discover current tool schemas and make a bounded first call.
Define proof before an agent says an action is done
Match user-facing claims to returned resources, provider acknowledgments and the measurements actually available.
Delegate scoped child keys without increasing authority
Choose workspace, project or inbox access and verify that a new credential stays within its parent’s boundary.
Design idempotent agent tool calls around one saved action
Keep operation identity stable across model continuations, worker restarts and uncertain network outcomes.
Manage multiple agent workspaces without confusing ownership
Keep workspace identity, human membership and resource credentials explicit when an operator serves several customers.
Preserve provenance when agents combine email, web and tool context
Keep user instructions, external claims and authoritative account facts separate in every proposal.
Read MCP results without mistaking transport success for tool success
Separate HTTP, JSON-RPC and operational failures, then report only the result the server established.
Recover agent registration with the original saved secret
Handle lost responses, conflicting requests and revoked credentials without creating replacement workspaces.
Register an agent through HTTPS with a recoverable identity
Create a bounded Free workspace and inbox without a human account, while keeping registration credentials recoverable.
Route agent tools through registered REST handlers
Keep one business authority for MCP, consumer agents and direct API clients instead of duplicating policy in each adapter.
Store agent keys and recovery secrets outside model context
Keep bearer tokens, registration identities and temporary control URLs private throughout provisioning and execution.
Understand an autonomous agent’s bootstrap permissions
Use reception, reading and drafts while keeping sending, key creation and paid effects outside the bootstrap boundary.
Use agent run traces to diagnose behavior without replaying actions
Inspect selected context, approvals and tool outcomes while respecting privacy, retention and measurement limits.
Use Lupin’s checked-in SDK and CLI without assuming public packages
Choose direct HTTPS or authorized source clients, then verify real operation coverage before building automation.
Validate MCP tool arguments against discovered JSON schemas
Use the concrete schema to reject unknown fields, missing identities and ambiguous provider versions before execution.