ApiSpi speaks both open agent protocols. Connect Claude Desktop, Claude Code, or any MCP client to one endpoint and it gets your governed connector tools. Or let another agent delegate whole tasks to Aria over Agent2Agent. Same tools, same governance, same key.
Your client's model does the thinking and calls your ApiSpi connector tools one by one — email, files, CRM, news, MCP servers you've added. Streamable HTTP, sessions included.
An external agent sends "summarise this week's customer emails" and gets the finished result back — Aria does the reasoning with your tools, under your policy.
Dynamic client registration and PKCE for clients like claude.ai, or paste a gateway API key from your profile. Both mint the same scoped credentials.
Blocked tools stay blocked, approvals still gate writes, and the emergency stop halts protocol traffic too — the policy doesn't care which client is calling.
Standard well-known documents for the agent card and OAuth metadata, so compliant clients configure themselves from a single URL.
Protocol traffic lands in the same usage logs, token accounting, and SIEM export as chat — one audit trail across every surface.
Works with Claude Desktop, Claude Code, and any MCP-compatible client