Setting the table

This is an honest comparison, so we will try not to oversell GeraNexus. MCP is great. A2A is great. The purpose of this post is to locate GeraNexus on the same map.

Model Context Protocol (MCP)

MCP was released by Anthropic in November 2024 as an open standard for connecting LLMs to tools, data sources and local context. It has become the de facto tool-calling protocol across the industry — within 12 months, first-party or community MCP support existed for every major model provider.

What MCP does well: read-side context. Your agent can query a database, search files, pull a calendar, access a file-system, call an internal service. It is the portable "tools menu" for an LLM.

What MCP does not do: it is not opinionated about write-side commerce. MCP tools can have side-effects, but the protocol does not standardise consent granularity, payment, escrow, policy exchange, dispute or verified completion. These are intentionally out of scope.

Agent-to-Agent (A2A) Protocol

A2A was announced by Google in April 2025 as a standard for agent- to-agent communication and task delegation. The canonical use case: agent A needs a capability agent B exposes; A delegates, B returns.

What A2A does well: orchestration. Coordinating multi-agent workflows, delegating subtasks, composing specialised agents into larger pipelines.

What A2A does not do: A2A assumes the downstream agents themselves understand commerce. It is a peer-to-peer messaging surface, not a commercial primitive.

GeraNexus

GeraNexus is deliberately a commerce protocol. It sits on top of MCP — most GeraNexus integrations are MCP servers with a thin wrapper that adds the transactional actions. The surface is: discover, negotiate, book, pay, verify-policy, confirm, verify- completion, dispute.

What GeraNexus does well: standardising the hard, boring stuff nobody wanted to standardise — consent scope, escrow state machine, policy exchange, dispute routing, completion receipts. It exists to turn "agent books a doctor" into a commodity integration.

What GeraNexus does not do: it does not replace MCP’s read-side tooling, and it does not try to orchestrate multi-agent workflows. If your agent wants to compose five specialist agents to book a holiday, use A2A to route the tasks and GeraNexus for each commercial leg.

How they stack

In a typical 2028-style agent request, you might use all three:

A2A orchestrates: "planning agent, please book a consultation" passes to a specialist health-booking agent.
MCP discovers: the specialist agent queries available providers via MCP-exposed directories.
GeraNexus transacts: the actual booking, payment and completion-verification flows through GeraNexus.

Protocol comparison

MCP: JSON-RPC 2.0, stdio or SSE transports, open spec, Anthropic-stewarded, broad adoption. A2A: HTTP-based, open spec, Google-stewarded, early adoption. GeraNexus: JSON-RPC 2.0 over HTTPS, OAuth 2.1 consent binding, Ed25519-signed receipts, escrow as a first-class primitive, open spec, Gera Services-stewarded, public drafting.

Where we see friction

Consent is the shared hard problem. MCP assumes the user has pre-authorised the tool; A2A assumes agents trust each other via some external mechanism. GeraNexus is the one that has to make consent explicit because it touches money. We are borrowing heavily from OAuth 2.1 and PAR (Pushed Authorization Requests), and probably introducing a minimal new primitive for consent-scoped-to-negotiation.

Are you competing with MCP or A2A?

No. MCP is the layer we sit on. A2A is a sibling layer. The world where agent commerce becomes common has all three protocols interoperating. We are deliberately in public about this to avoid standards fragmentation.

Learn more

The GeraNexus draft spec lives at /spec, open questions go up on this blog, and the waitlist is for integrators and researchers who want to help shape the v1. Expect us to be corrigible in public.