Site search

Find architecture, research, and terms

Start typing to search the editorial index.

LandscapeTechnical

Protocols and Ecosystem

How MCP, A2A, AI gateways, model servers, workflow engines, identity, observability, memory stores, and sandboxes relate to MIR.

MIR is not a replacement for every adjacent AI infrastructure category. It is the execution-control layer that composes model serving, protocols, gateways, workflow systems, identity, storage, and isolation around a governed run.

Key takeaways

  • Protocols define communication; the runtime defines authority and lifecycle.
  • Gateways manage traffic; MIR manages task state and side effects.
  • Observability records operations; evidence explains a run to users and reviewers.

Applications and users
Machine Intelligence Runtimestate · authority · tools · memory · recovery · evidence
MCPtools, resources, prompts
A2Aagent interoperability
AI gatewaysmodel traffic and routing
Workflow enginesdurable transitions
Identity and vaultsauthentication and credentials
Observabilitytraces, metrics, logs
Memory storesvector, graph, relational, events
Sandboxesprocess, filesystem, egress
Model servers and inference runtimes → hardware
Neutral conceptual map. Categories overlap in products, but their responsibilities remain distinct.

MCP placement patterns

MCP standardizes how a host discovers and invokes tools, resources, and prompts. It can be placed locally, behind a gateway, inside an isolated runtime, or across a managed boundary. The host remains responsible for consent, authorization, lifecycle, validation, and evidence. Source: MCP specification

Agent interoperability

A2A targets communication between independently built agent systems. In a MIR design, incoming agent messages are untrusted requests that need identity, policy, scope, budgets, and evidence. A protocol can carry a task or artifact; it cannot establish organizational trust by itself. Source: A2A project

Adjacent systems

Category Primary contribution MIR integration concern
AI gateway Model routing, quotas, traffic policy Run identity, task budget, evidence correlation
Model server Inference, batching, cache, hardware Capability contract, model version, uncertainty
Workflow engine Durable deterministic transitions Stochastic validation and context governance
Identity provider and vault Authentication and secret issuance Dynamic least privilege for each action
Memory stores Vector, graph, relational, event persistence Scope, provenance, consent, correction, deletion
Sandbox or microVM Process and filesystem isolation Image trust, egress, cleanup, artifact capture
OpenTelemetry Traces, metrics, logs, semantic conventions User-facing evidence and privacy overlays

Selection questions

  • Which component owns identity and the final authorization decision?
  • Where is runtime state durable, and who can resume it?
  • How are tool schemas, permissions, and versions distributed?
  • What evidence survives provider or framework replacement?
  • Can the organization stop, isolate, export, and delete a run?

Source record

References

Suggest a correction
  1. Model Context Protocol project. Model Context Protocol. Published 2025-11-25; last reviewed 2026-06-24 UTC. Protocol specification.

  2. A2A Project. Linux Foundation. Published Current specification; last reviewed 2026-06-24 UTC. Protocol specification and reference implementation.

  3. OpenTelemetry project. Cloud Native Computing Foundation. Published Current specification repository; last reviewed 2026-06-24 UTC. Official specification.

  4. Microsoft and ONNX Runtime contributors. ONNX Runtime. Published Current documentation; last reviewed 2026-06-20 UTC. Official documentation.

  5. vLLM project. vLLM. Published Current documentation; last reviewed 2026-06-20 UTC. Official documentation.