Where Saakshya sits next to W3C credentials, MCP and A2A

A woman asks her assistant to get her a small loan for a laptop. The assistant needs her last three salary slips, so it uses a tool that can read her documents folder. Then it asks a second agent, one that knows how to fill in bank forms, to prepare the application. That agent sends the finished form to a bank.

Most steps in that story can already run on a shared standard. The assistant reached the document tool in a standard way. The two agents spoke to each other in a standard way.

Now look from the bank's side. A form arrives. The bank's officer wants to know who is answerable for this. Not how the message got there. Who.

That's the gap we work on, and it's why we get asked how Saakshya relates to the standards people already know.

The standards, one at a time

W3C credentials. W3C is the body that publishes the web's open standards. Its Verifiable Credentials Data Model v2.0 became a W3C Recommendation on 15 May 2025 (page read 12 October 2026). It describes a standard way to record claims that an issuer makes about a subject, in a form that can be checked for tampering, such as a licence or a certificate. W3C's announcement of the same date calls the Verifiable Credentials 2.0 family a W3C Standard.

W3C identifiers. Decentralized Identifiers (DIDs) v1.0 became a W3C Recommendation on 19 July 2022. The page describes a DID as an identifier that its controller can create and prove control of. A newer DID v1.1 was a Candidate Recommendation Snapshot dated 5 March 2026 when we read it on 12 October 2026, so it isn't final as far as we can tell.

MCP. The Model Context Protocol describes itself as an open-source standard for connecting AI applications to external systems, such as data sources and tools (read 12 October 2026). Anthropic introduced it on 25 November 2024. On 9 December 2025 Anthropic announced that it was donating MCP to the Agentic AI Foundation, which the post calls a directed fund under the Linux Foundation, and that MCP's governance model would stay unchanged.

A2A. The Agent2Agent protocol is described on its own site as an open standard for communication between AI agents. The site says Google originally developed it and donated it to the Linux Foundation. The Agentic AI Foundation's post of 17 August 2026 says A2A is joining that foundation as a hosted project, and says agents publish cards describing their capabilities so other agents can find them.

What they leave open

Read the descriptions again. MCP is about how an agent reaches tools and data. A2A is about how agents find each other and work together. Both are about the connection.

Neither, as far as we can tell from those pages, sets out to tell an office who is answerable for an agent that turns up. An agent card says what an agent can do. It doesn't say which company or person will pick up the phone if the loan form is wrong. Both projects may well have security and authorisation material that we haven't read in full, and that would be about the connection too.

W3C's credentials are different. They're a format, and a good, open one. They can carry a claim, but a format doesn't say who is allowed to make the claim or whom an office should believe. That needs someone to do the registering and checking.

Where we sit

In the picture, the W3C standards are the base. Saakshya sits in the middle, with MCP on one side and A2A on the other.

We're designing Saakshya to work with W3C's standards for digital credentials, because we'd rather not invent a format when an open one exists. We mean to sit alongside MCP and A2A and not compete with them. An agent can use MCP to reach a tool and A2A to ask another agent for help, and still need an identity that a bank can check.

That's all we're saying. We're not claiming that Saakshya has been built into any of these, that any of them has adopted it, or that anyone at W3C, the Linux Foundation, Anthropic or Google has looked at our work. We haven't spoken to them, and none of them has endorsed us. The foundation isn't incorporated yet.

If you want the longer argument for why a register has to sit outside any one company, we made it in an earlier piece. And we wrote about what the protocols leave to each vendor in Interoperability is standardising. Retention is not.

What we don't know

We don't know how closely the W3C standards will fit what an Indian bank or hospital wants to check. They were written to be general. We think that's a strength, and we haven't tested it against a real office.

We don't know where the DID standard will settle. The newer version was still a candidate when we looked, and it could change before it's final.

We don't know whether the people who run MCP and A2A will want an identity layer of their own. The Agentic AI Foundation's site lists an Identity & Trust working group (read 12 October 2026), described as covering portable identity and how permissions flow between agents, so the question isn't being ignored. If a good answer comes from there, we'd rather work with it than race it.

And we've read these pages on one day. Standards move, hosting moves, and what we've said about who looks after which project is a snapshot from 12 October 2026.