Skip to content

CONCEPT Cited by 1 source

Multi-surface MCP

Definition

Multi-surface MCP is an architectural pattern where a single agent's tool catalog is composed from multiple Model Context Protocol (MCP) surfaces with different latency, sharing, ownership, and scaling properties — presented to the agent as one unified catalog.

Instacart's Blueberry uses three surfaces:

  1. In-process MCP — fast local helpers that share session state with the agent runtime. Low latency, tightly coupled.
  2. Standalone shared MCP server — heavier investigators (deploy analysis, service-dependency lookup, error analysis, anomaly sweeps). Can be scaled and updated independently from the agent runtime. Shared across teams.
  3. Team-hosted remote MCPs — mounted remotely so team-unique data sources appear as first-class tools without being baked into the core. Different ownership boundary.

"The three surfaces have different latency, sharing, and ownership properties, and the processor presents them to the agent as one tool catalog." (Source: sources/2026-07-14-instacart-blueberry-on-call-reasoning-harness)

Design rationale

  • Latency stratification: local helpers are fast; shared investigators can be slower but more capable
  • Independent scaling: heavy investigators don't need to be co-deployed with every agent worker
  • Ownership boundaries: teams can own and iterate on their MCP endpoints without touching core
  • Unified agent experience: from the agent's perspective, the right tools are simply present

Seen in

Last updated · 585 distilled / 1,765 read