Claude.ai (browser/app)
Claude.ai in web browser or mobile app. Runs in Anthropic's hosted environment. Code execution and file operations are session-sandboxed.
Agent Passport
APAI.agent-passport.v0.1Agent Passport
cloud-sandboxunverified v0.1
Agent: Claude.ai (browser/app)Vendor: AnthropicVersion: varies (hosted, rolling)
Install modes:cloud-sandbox
↓
Can read
- ·User-supplied text in the conversation
- ·Files uploaded by the user for the session
- ·Project knowledge uploaded by the user to a Project
- ·Content fetched via custom Connector (MCP) when user has configured one
↑
Can write
- ·Session sandbox only (Artifacts: code, markdown, SVG, React previews - shareable link but not persisted on user's machine)
- ·Code execution output in session sandbox (not written to user's filesystem)
↔
Can access
- ·Anthropic Claude API for inference
- ·User-configured custom Connectors (MCP servers added via claude.ai/connectors)
- ·Computer Use in select plans (desktop control within a remote sandbox, not user's local machine)
$
Can spend
- ·User's Claude.ai subscription quota (message limits per billing period)
- ·No direct money transfer initiated by the agent
○
Can expose
- ·Conversation content sent to Anthropic API for inference
- ·Files and project knowledge included in API context
- ·Connector tool results returned to Claude include data from the connector source
✓
Will request approval for
- ·Connector actions that require user confirmation per connector's declared schema
- ·Artifacts are shown inline; user chooses to copy or download
v0.1 note: This passport is descriptive, not enforceable. Runtime enforcement is the Policy Pack + Gateway story (Phase 4+). Fields sourced from publicly documented agent behavior as of May 2026.
Compatible packages (v0.1)
APAI packages that declare support for hosted chat platforms compatible with this agent. v0.1 match is by platform overlap; full capability-field matching ships with Phase 4.