[Headroom Vercel SDK](https://headroom-docs.vercel.app/docs) Headroom’s Compress Cache Retrieve pattern as part of a MCP server ```mermaid sequenceDiagram participant Client participant MCP as MCP Server participant Headroom participant CCR as Context Store participant Retrieval as CCR Retrieval participant LLM %% Initial request Client->>MCP: Send request MCP->>Headroom: Forward full context: `compress` %% Compression + storage Headroom->>CCR: Store full context Headroom->>LLM: Send compressed prompt %% LLM asks for more info LLM-->>Headroom: Tool call (need more data) Headroom-->>MCP: Forward tool request %% Retrieval flow (now explicit) MCP->>Retrieval: Invoke retrieval tool `ccr_retrieve` Retrieval->>CCR: Fetch full context CCR-->>Retrieval: Return data Retrieval-->>MCP: Return relevant chunk %% Retry with more context MCP->>Headroom: Send enriched prompt Headroom->>LLM: Resend request LLM-->>Headroom: Final response Headroom-->>MCP: Forward response MCP-->>Client: Final answer ```