Concept overview
The 5-Contract Architecture
Every agent in Multipliers — built-in, third-party, or Foundry-authored — must satisfy five runtime contracts before it can be published into a tenant’s active fleet. The contracts are structural: they’re enforced at the publish gate, not reviewed in a PDF.
We chose five contracts because each one addresses a recurring failure mode we’ve seen in enterprise AI deployments — opaque tool surfaces, agents that can’t federate, outcomes that don’t roll up to the CFO, missing audit lineage, and process bindings that drift from reality.
What the full guide will cover
- Contract 1: MCP server — every agent exposes its tools via Anthropic MCP for inspection and external invocation
- Contract 2: A2A AgentCard — every agent is discoverable and callable through the Agent-to-Agent protocol
- Contract 3: MVE outcomes — every run emits a Multipliers Value Envelope outcome that rolls up to the value-chain dashboard
- Contract 4: Context Engine integration — every agent reads/writes through the Context Engine layer for cross-source grounding
- Contract 5: Value-chain bindings — every agent binds to a named process step in the enterprise process graph
- The publish gate — shadow → testing → active, with ≥90% match rate required before live writes
- The compliance dashboard — per-agent status of all 5 contracts at /governance
Coming soon — request the full guide
The full 5-Contract reference is being authored with worked examples from each shipping agent. Email the FDE team for early access.
Request full guide