There is a quiet problem growing at the centre of the AI agent ecosystem.

Thousands of MCP servers are live. Agents are calling tools, delegating tasks, and passing access tokens across systems at a pace that nobody fully anticipated even two years ago. The infrastructure largely works. But if you ask the question “how does anyone actually verify the identity and reputation of an agent they’ve never interacted with before?” the answer gets uncomfortable… fast.

It’s not that the identity layer doesn’t exist. It’s that the trust layer on top of it is thin. Difficult to audit and to verify independently.

That’s the problem cheqd and Vouched’s MCP-i framework, recently renamed to KYA-OS, are solving together.

 

The Setup: 27,000 Servers and a Working Identity Model

Vouched, the team behind KYA-OS and the KnowThat.ai registry, already has a working identity layer for MCP servers built around did:web. Around 27,000 MCP servers and agents are currently registered under did:web:knowthat.ai:*. The middleware handles signed proofs, delegation, and access verification. It works.

Flipping that to a new identifier system overnight would break every consumer relying on it. That’s obviously not the goal and not what this integration proposes.

The KYA-OS protocol, as documented in the kya-os middleware layer, already handles Ed25519/EdDSA signatures end to end, with a clear roadmap toward post-quantum cryptography being developed in parallel through the DIF Task Force. The signature stack is compatible and the identity model is sound. What’s missing is an independently verifiable layer for trust metadata and reputation.

That’s where cheqd comes in.

 

The Approach

The proposed architecture is straightforward in principle, even if the implementation has detail worth understanding.

The existing did:web identifiers stay. Nothing breaks for existing deployments. A corresponding did:cheqd identifier is created alongside the existing one. The two are linked using alsoKnownAs in a bidirectional model, because a one-way claim wouldn’t be independently verifiable. A resolver that can only confirm the claim in one direction can’t tell a real link from a hijacked one. New registrants are expected to default to did:cheqd over time.

From there, cheqd’s DID-Linked Resources (DLRs) provide the anchoring layer for selected persistent artefacts. What gets anchored is what genuinely benefits from persistence and independent discoverability.

Capability manifests, conformance levels, trust and reputation score snapshots, configuration and delegation metadata. Information that a third party would actually want to verify without having to trust your database.

 

The Part That Could Change the Trust Registry Story

The most interesting thread in the technical alignment notes between the two teams concerns reputation scores.

KYA-OS currently produces deterministic reputation scores that are DID-subject-bound but delivered as plain JSON. There are no signatures on them. Here’s the opportunity to make the trust layer independently verifiable.

Wrapping those score snapshots as Verifiable Credentials and anchoring them as DLRs under the agent’s did:cheqd changes that entirely. A third party would be able to derive the score from the published inputs, verify the signature, and reach the same number without touching Vouched’s database. The methodology is transparent. The output is independently checkable.

For the AI agent ecosystem specifically, this matters. The question of whether an agent is trustworthy enough to call a tool or receive a delegation is currently answered informally. A verifiable, anchored reputation score with a clear input hash and signing key changes that from a social trust question to a cryptographic one.

 

The Implementation Path

The integration is being structured as an adapter first, rather than a direct upstream contribution into the KYA-OS codebase. This keeps governance overhead out of the critical path. The plan is to publish the work as canary packages under the existing @kya-os/mcp package name, share the approach with the DIF TAAWG working group, and let the generally useful patterns (alsoKnownAs verification, the DLR usage patterns, status list integration) flow upstream through the working group on their own timeline.

The direction is for did:cheqd to become canonical for new registrants into KnowThat.ai, and for existing servers to migrate to did:cheqd as canonical once they go through a claim flow. identities with unmatched speed, accuracy, and regulatory compliance–accelerating trust in an increasingly digital world.

 

What This Is Really About

The AI agent ecosystem is still working out what accountability looks like at scale. The tools exist. The identity layer exists. What’s missing is a trust and reputation layer that survives a handoff between parties who don’t already know each other.

The commercial pressure to solve this is building fast. Mastercard’s Verifiable Intent, announced March 2026, is an open-source, standards-based framework designed to link a consumer’s identity, their specific instructions, and the outcome of a transaction into a single tamper-resistant record. The framing from Mastercard’s own CDO is pointed: ‘As autonomy increases, trust cannot be implied. It must be proven.’

Meanwhile, Google’s Agent2Agent protocol, now under Linux Foundation governance with 150+ organisational supporters, provides the interoperability layer for agents operating across organisational and vendor boundaries. Both frameworks point toward the same unsolved problem: agent identity that is independently verifiable, not just self-declared.

How do we build trust in this space? By leaning into open standards and ensuring no single organisation controls how agents are rated or recognised. And by making that data accessible without requiring a commercial relationship to reach it.

cheqd’s DID infrastructure, combined with the KYA-OS and Knowthat.ai’s conformance and reputation model, is a credible architecture for that. It has a defined first implementation scope, a clear view of what belongs on-chain versus off, and a path to becoming a genuine interoperability standard through DIF, the kind of infrastructure that frameworks like Verifiable Intent and A2A will need as they mature.

 

Get in touch

If you are building MCP servers, operating AI agent infrastructure, or working on trust and identity in the agent ecosystem, we’d like to talk.

The architecture described above is open for review, and we are actively looking for operators who want to build with us and help shape the conformance and reputation model.

Get in touch with the cheqd team and reference KYA-OS in your message. We will get back to you directly.