Google's A2A Protocol has a trust gap: developers integrating agents built on the spec have no standard way to verify one actually works before wiring it into production. A2Apex Ventures LLC launched a public beta this week to fill that gap, with a testing and certification platform that runs automated compliance checks against the A2A specification and publishes the results as a scored, searchable record.
The platform validates agent cards, tests live JSON-RPC endpoints, and checks security configurations including API keys, OAuth, and JWT. Agents receive a 0–100 trust score that maps to Gold, Silver, or Bronze certification badges, along with a public profile page in a directory other developers can search before integrating. Pricing runs from a free tier — five tests per month, badges expiring after 90 days — up to $499 per month for unlimited tests, permanent badges, and CI/CD integration.
The certification-plus-discovery combination targets a genuine problem. Frameworks like LangChain, AutoGPT, and <a href="/news/2026-03-14-gitagent-an-open-standard-for-turning-git-repos-into-ai-agents">CrewAI</a> don't natively interoperate, and no standardized mechanism exists for verifying agent reliability before deployment. Bundling both certification and discovery into one platform drew notice from commenters on Hacker News, who flagged it as a differentiator from tools that tackle only one side of the problem. A CLI and SDK (pip install a2apex) are on the roadmap for Q2 2026, with an agent marketplace featuring discovery APIs and premium listings planned after that.
The single-protocol bet carries real risk. A2A launched at Google Cloud Next in April 2025 with 50-plus partner companies, then moved to the Linux Foundation in June 2025 under a multi-stakeholder structure with founding members including AWS, Microsoft, Cisco, and Salesforce. The spec is Apache 2.0 licensed and currently at v0.3. Linux Foundation governance lowers the odds that any one steward formalizes a competing certification program and renders A2Apex redundant — but it doesn't rule it out. As of March 2026, neither Google nor the Linux Foundation's A2A project has announced a verification program, which is the gap A2Apex occupies.
Protocol competition adds a second layer of exposure. Anthropic's Model Context Protocol (MCP), also now housed under the Linux Foundation's Agentic AI Foundation, has hit 97 million monthly SDK downloads and runs across ChatGPT, Claude, Gemini, and Microsoft Copilot. MCP and A2A serve adjacent rather than identical purposes, but a compliance suite built entirely around one protocol means any major shift in industry alignment would force a rebuild from the ground up.