Four agent protocols now compete for adoption, and NIST has opened an initiative to align them. Each one solves transport, discovery and the call contract, and each one solves those well. None of them tells the receiving system what a field means. The mapping work therefore survives the protocol, and this article shows exactly where it lands.
Connect with us about connected systems · · Replies within 24 hours
No. The Model Context Protocol gives a model a standard way to find a tool, call it, and receive a result. It defines the envelope, and it does not define the contents.
A tool can describe itself as returning a customer record with a status field. The protocol carries that description faithfully. It does not say which values the status field takes in your organisation, or which of them count as active for a report.
Transport and meaning are separate problems, and a protocol solves the first one.
| Protocol | What it standardises | Does it carry business meaning? |
|---|---|---|
| MCP | How a model reaches a tool or a data source. | No. |
| A2A | How one agent discovers and calls another. | No. |
| ACP | Agent communication, with a focus on messaging. | No. |
| ANP | Agent discovery and interaction across networks. | No. |
The A2A specification covers the second row. A published survey of agent interoperability protocols compares all four.
The fourth column holds the same answer four times, and that repetition is the finding. No protocol claims to solve meaning, and each one is honest about its scope. The claim appears in the marketing that follows.
In the integration code, once per pair of systems. Consider two systems that both hold a customer.
System A calls it a client with a status of live, pending or closed. System B calls it an account with a state of 1, 2 or 3. A protocol moves both records reliably, and somebody must still write the rule that says live equals 1.
That rule sits in code between the two systems. Where a third system arrives, you write two more rules. Where a fourth arrives, you write three more.
Nothing visible. A mapping error passes a value that the receiving system accepts, so no error appears and the data is now wrong.
Three properties make this the expensive failure mode.
NIST announced its AI Agent Standards Initiative in February 2026, to work towards interoperable and secure agents.
Recent research examines what the protocols leave open, in governance gaps in agent interoperability protocols.
Standardisation helps. One protocol instead of four reduces your integration surface, and it does not change what a status field means in your organisation. That fact is local to you, and no standards body can publish it.
A 2026 benchmark measured it. Three frontier models answered 99 business questions on a public dataset, twice each. The second run added a 4KB document describing the measures and the conventions.
Accuracy rose by 17 to 23 percentage points for every model. With the document present, the three models became statistically indistinguishable from each other.
The document supplied meaning, and no protocol supplies that. We give the full evidence in why AI gets your own business data wrong, and the vocabulary in semantic layer, ontology or knowledge graph.
We build your system on engage.re, and the platform removes the mapping rather than moving it.
The platform holds everything in five database tables, and those five never change. Sense Future built engage.re, and it has run in production since December 2025.
Yes, and for the reason it offers rather than the one the marketing offers.
Gartner expects 75% of API gateway vendors to carry MCP features by 2026, so the transport question is settling. The meaning question stays with you.
A council connects housing, revenues and adult social care through a standard protocol. The three systems exchange records reliably, and each one holds a different definition of a household. The protocol delivered the data and the reports still disagree.
A care home group connects a records system, a rota tool and a family portal. Three systems produce up to three mappings, and the group maintains all three. One of them defines an occupied bed by admission and another by invoice, so the occupancy figure depends on which system asked.
The care home software and school software pages set out the sector detail. Our guides to charity data management and multi-academy trust systems show the same disagreement inside one organisation.
No. MCP gives a model a standard way to find a tool, call it and receive a result. It defines the envelope and not the contents. A tool can describe a status field, and the protocol will not say which values count as active in your organisation.
MCP standardises how a model reaches a tool or a data source. A2A standardises how one agent discovers and calls another. Both solve transport, discovery and the call contract, and neither carries business meaning.
Into integration code, once per pair of systems. One system may call a record a client with a status of live, and another may call it an account with a state of 1. Somebody must write the rule that says live equals 1. Ten systems need up to 45 such mappings.
Nothing visible. A wrong mapping passes a value the receiving system accepts, so no error appears and the data is now wrong. The discovery usually comes months later from a person who notices a strange figure, and nobody can bound the affected records without checking them all.
Yes, for the connector work it saves. Adopt MCP to give models a standard route to your tools, and A2A where agents must find each other across boundaries. Do not expect either to tell a receiving system what your fields mean, and record your meaning separately.
We build your system on engage.re, which removes the mapping rather than moving it. Every application uses one shared dictionary with deterministic identifiers, so your systems agree by construction. The server enforces your declared vocabulary, and a wrong meaning fails at the write instead of surfacing in a report months later.
Connect with us about connected systems · · Replies within 24 hours