The Claude Platform Glossary is a collection of terms used in Claude Platform, Anthropic's developer API surface for Claude, spanning the Messages API, Claude Managed Agents, MCP integrations, and the broader programmable surface.
Glossary
A
- Advisor tool — beta Claude Platform capability that lets a model consult a larger or more capable Claude model mid-task for a second opinion or harder reasoning.
- 📝Agent Loop — the orchestration cycle inside Claude Managed Agents: model picks a tool, harness executes it, result returns to context, model decides the next step. Runs on Anthropic's infrastructure even when sandboxes are self-hosted.
- Agent Skills — beta Claude Platform feature exposing reusable, file-based capabilities to agents, mirroring the Skill model in 📝Claude Code.
C
- Claude Console — the web app at platform.claude.com for managing workspaces, API keys, prompts, evals, and Managed Agents.
- 📝Claude Managed Agents — Anthropic's hosted multi-step agent runtime. Currently in beta. Priced per token plus $0.08 per active session-hour.
- Claude Platform on AWS — generally available service launched May 2026 that delivers the native Claude Platform experience through an AWS account, with IAM auth, CloudTrail logging, and AWS Marketplace billing. Same pricing as direct.
- Cloud Container — Anthropic-managed sandbox environment that executes tool calls inside Managed Agent sessions by default. Replaced by Self-Hosted Sandboxes when self-hosting is enabled.
- Code Execution — Claude Platform feature for running Python in Anthropic's managed sandbox, used for data analysis and visualization directly inside API calls.
E
- Environment Worker — the process running inside a customer's infrastructure that claims work items from Anthropic's queue, executes tool calls, and returns results. Core component of Self-Hosted Sandboxes.
F
- Files API — beta Claude Platform endpoint for uploading and referencing files across requests, used for documents, images, and code artifacts.
M
- MCP Connector — beta Claude Platform feature that connects remote MCP servers directly to Messages API requests without a local client.
- 📝MCP Tunnels — research-preview Claude Platform capability that lets Managed Agents reach private MCP servers inside a customer's network through a single outbound encrypted gateway connection, without exposing the server publicly.
- Messages API — Claude Platform's primary endpoint at
/v1/messagesfor chat-style completions, function calling, streaming, batch processing, and prompt caching. - 📝Model Context Protocol (MCP) — the open standard for exposing tools, resources, and prompts to LLM agents. Used by Claude Platform via MCP Connector, MCP Tunnels, and direct integrations.
- 📝MCP Server — process implementing MCP that exposes tools to Claude Platform clients over stdio or HTTP.
P
- Prompt caching — Claude Platform feature that caches tools, system prompts, and message history with a ~5-minute TTL to reduce latency and cost on repeated context.
S
- 📝Self-Hosted Sandboxes — public-beta Managed Agents feature that moves tool execution from Anthropic's Cloud Containers to a customer-controlled sandbox runtime, either custom-built or hosted via partners (Cloudflare, Daytona, Modal, Vercel). Not yet available on Claude Platform on AWS; Memory is not yet supported.
- Session-hour pricing — the $0.08 per active runtime hour billed on top of standard token rates for Managed Agent sessions.
W
- Web fetch — Claude Platform tool that retrieves the contents of a specific URL on the model's behalf.
- Web search — Claude Platform tool that runs a search engine query and returns ranked results to the model.
- Workspace — the top-level organizational unit in Claude Console for separating projects, environments, or teams while maintaining centralized billing and admin.