FOR MCP PROVIDERS · EU AI ACT ARTICLE 50 · IN APPLICATION SINCE 2 AUGUST 2026
Your MCP isn't automatically in Article 50 scope. The agents using it may be — and buyers will ask either way.
Running an MCP server does not automatically place a provider in Article 50 scope. The AI systems and agents that use it may carry transparency obligations, and enterprise buyers may still require accountability evidence about the agent, its operator and the authority it carries. ECZ-ID makes that evidence portable and re-checkable — so those questions stop stalling deals.
See the Agent Evidence pilot →- Purchasable plans from Free · MCP Assurance from £199/month
- 8-page executive briefing · free · no email required
2 August 2026
In application
Article 50 applies
Transparency obligations for providers and deployers are in application (Article 113). Some enterprise customers may be in scope now.
2 December 2026
Deadline
Article 50(2) transition ends
Deadline for qualifying generative systems placed on the market before 2 August 2026 to meet the machine-readable marking obligation (Article 111(4), inserted by Regulation (EU) 2026/1744).
The Code of Practice on Transparency of AI-generated Content supports implementation of the marking duties — following a code of practice is not a finding of compliance.
What your enterprise customers are about to ask you
Article 50 made AI accountability a board-level topic at exactly the moment agents started calling third-party MCP surfaces in production. Some of your enterprise customers may carry Article 50 duties they need to evidence — and many more will ask accountability questions whether or not Article 50 applies to their particular use case.
“Which agent is calling our MCP, and who operates it?”
Answered ad hoc in a spreadsheet, re-asked at every renewal.
“Whose authority does that agent carry?”
Usually unanswerable once an integrator sits in the middle.
“What transparency evidence stands behind the operator?”
A PDF attachment that was stale the day it was emailed.
“Can we re-check any of this ourselves, later?”
Today: no. So diligence repeats in full, every time.
None of that is a protocol problem, so no amount of gateway work fixes it. It is an evidence problem — and evidence is portable, which means it can be solved once and reused by every customer who asks.
MCP provider plans
Purchasable products with published prices, operated through TrustOps. Start free, and add ongoing Resolver proof when a customer asks you to prove something twice.
ECZ-ID MCP Verifier™
Free
Resolve a Passport, inspect what a page actually asserts, and see the basis of each field.
Anyone checking a provider before they rely on it.
ECZ-ID MCP Assurance™
£199/month
Keep your MCP provider identity and evidence current and independently resolvable, with the lifecycle operations that hold it accurate as your surfaces change.
Providers whose customers re-check them.
Review & buy →ECZ-ID MCP Assurance Plus™
£499/month
Extend that evidence posture across broader MCP surfaces and targets, within the entitlements the canonical product defines.
Providers operating several surfaces or targets.
Review & buy →ECZ-ID MCP Policy™
£1,499/month
Publish machine-readable buyer-side acceptance policy for MCP surfaces — what your organisation will and will not accept from what it consumes, stated under your own authority.
Buying organisations consuming third-party MCP surfaces.
Requires the Assured parent. The Assured parent entitlement is included in this £1,499/mo price.
Review & buy →Acquisition, billing and lifecycle for these products are handled by TrustOps. MCP Assurance, Assurance Plus and MCP Policy all include the parent entitlement they require in the price shown, so there is no separate Business Passport line for any of them. MCP Policy is the one that requires the Assured parent rather than the Verified one. These are monthly products, not the fixed-scope engagement further down this page.
The MCP Verifier is a free machine-readable checker. There is no paid checkout for it.
Agent Credential and operator accountability (KYA)
The evidence your customers actually want is about the agent, not the endpoint: which agent this is, who runs it, and on whose authority it acts. These are purchasable products with published prices.
Agent Credential
An agent identity issued as a child of an accountable parent ECZ-ID, so an agent always resolves back to an organisation that can be held to account.
Operator & principal binding
Who operates the agent, and whose authority it carries — the two questions an integrator in the middle usually makes unanswerable.
Resolvable, not asserted
The relying party re-checks the record at the moment of reliance instead of trusting an attachment. Absence resolves as absence, truthfully.
ECZ-ID Agent Credential Entry Pack™
£24.99/mo
Give an agent a persistent, resolvable identity linked to its accountable organisation.
Agent identity. For operator evidence, start at KYA Starter.
Review & buy →ECZ-ID KYA Starter™
£34.99/mo
Publish operator and principal accountability evidence for agents — who runs this agent, and on whose authority it acts.
First Know-Your-Agent evidence for a live agent.
Review & buy →ECZ-ID Verified KYA Ready Pack™
£149.97/mo
The Verified-tier operator evidence posture, for providers whose buyers run real diligence.
Providers answering procurement questionnaires.
Review & buy →ECZ-ID Assured KYA Ready Pack™
£179.97/mo
The Assured-tier posture, for relationships where the relying organisation's bar is highest.
Regulated or high-assurance counterparties.
Requires the Assured parent. The Assured parent entitlement is included in this £179.97/mo price.
Review & buy →The Entry Pack, KYA Starter, the Verified pack and the Assured pack all include the parent entitlement they require in the price shown, so there is no separate Business Passport line for any of them. The Assured pack is the one that requires the Assured parent rather than the Verified one. Nothing here is required by Article 50; these exist because relying parties ask.
Why this compounds
Evidence has a property questionnaires do not: it is reusable. Each turn of this loop lowers the cost of the next enterprise conversation — for you and for the agents on your surface.
01
You publish evidence once
Identity, operator and supporting evidence, each with its stated basis and review date.
02
Buyers re-check instead of re-asking
A resolvable record answers the same question for every customer, without another questionnaire round.
03
Make your MCP easier to review
Portable, re-checkable evidence can reduce repeated diligence and give reviewers clearer answers without another questionnaire round.
04
Agents on your surface are invited to publish too
As the layer matures, your platform becomes a distribution point for accountability rather than a dead end for it.
Step 04 describes the design-partner direction rather than shipped behaviour. It is available now as a pilot engagement — see the Agent Evidence section below.
What re-checkable evidence actually looks like
The shape of a resolvable evidence profile: every field carries its basis, its state and its review date — and anyone relying on it can re-check it instead of filing it.
Every field carries its basis, its state and its review date, and the same record is readable by a person and by a machine. Nothing here is a score, a rating or an approval.
ECZ-ID EVIDENCE PROFILE
Illustrative example
Synthetic identifiers. Not a live record.
- AI system / agent
- Customer support agent “Aria”
- Identity
- agent:ecz:demo-a50-0001
- Operator
- Example Retail Ltd
- Principal
- Example Retail Ltd
- Article 50 evidence
- Interaction disclosure · customer-declared
- Disclosure evidence
- First-interaction notice captured
- Provenance
- Output marking method recorded
- Last reviewed
- 5 August 2026
- Current state
- Published · re-checkable
Machine-readable
{ "agent": "ecz:demo-a50-0001", "state": "published", "reviewed": "2026-08-05" }Resolver reference
resolver → read-only proof projection
Article 50 in sixty seconds
Since 2 August 2026, Article 50 of the EU AI Act (Regulation (EU) 2024/1689) applies transparency obligations to specific roles in specific situations. They are separate obligations — most organisations carry some of them, many carry none, and almost nobody carries all of them at once.
Article 50(1)
Providers
Tell people when they are talking to an AI system
AI systems intended to interact directly with natural persons must be designed so those persons are informed they are interacting with an AI system — unless that is obvious to a reasonably well-informed, observant and circumspect person in the circumstances, or a narrow legal exception applies.
Article 50(2)
Providers
Mark synthetic content in a machine-readable way
AI systems, including general-purpose systems, that generate synthetic audio, image, video or text must mark their outputs in a machine-readable format so the content is detectable as artificially generated or manipulated — with technical solutions that are effective, interoperable, robust and reliable as far as technically feasible, and with limited exceptions such as assistive editing.
Article 50(3)
Deployers
Disclose emotion recognition and biometric categorisation
Organisations deploying an emotion-recognition or biometric-categorisation system must inform the natural persons exposed to it about its operation, subject to narrow law-enforcement exceptions.
Article 50(4)
Deployers
Disclose deep fakes and certain published AI text
Deployers must disclose that deep-fake image, audio or video content has been artificially generated or manipulated, and must disclose AI-generated or manipulated text published to inform the public on matters of public interest — with exceptions, including where the text has undergone human review and a person holds editorial responsibility.
Article 50(5)
Both
Make the disclosure clear, timely and accessible
The required information must be provided to the natural persons concerned in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure, and in line with applicable accessibility requirements.
Limited transition: Providers of qualifying generative systems placed on the market before 2 August 2026 have until 2 December 2026 to take the necessary steps to comply with the Article 50(2) marking obligation. The other Article 50 obligations carry no such transition.
Does Article 50 apply to me?
Article 50 attaches to situations, not to sectors — so the honest answer is an orientation, not a verdict. Check your role and use case against these four situations.
Interactive AI
Article 50(1) may apply
Chatbots, voice agents and assistants intended to interact directly with natural persons. If a person could reasonably not realise they are talking to a machine, check your role — the provider carries the disclosure duty.
Generative synthetic content
Article 50(2) may apply
Systems generating synthetic audio, image, video or text. Providers should check the machine-readable marking obligation — and the 2 December 2026 transition if the system was on the market before 2 August 2026.
Emotion recognition / biometric categorisation
Article 50(3) may apply
Deployers of these systems must inform the people exposed to them. The duty lands on the deploying organisation, not the software vendor.
Deep fakes / public-interest AI text
Article 50(4) may apply
Deployers publishing deep-fake media, or AI-generated text that informs the public, carry disclosure duties — with exceptions for editorially controlled text and limited artistic contexts.
Purely internal tooling, systems whose AI nature is obvious in context, and background machine-to-machine integrations do not automatically attract Article 50 duties. “Not assessed” is an honest interim state — “out of scope” is a conclusion worth documenting, not assuming.
For completeness, once: breaches of the Article 50 transparency obligations sit under Article 99(4)(g), with administrative fines up to €15 million or 3% of total worldwide annual turnover, whichever is higher — and whichever is lower for SMEs. Those are ceilings set by Member State authorities against statutory proportionality factors, not expected outcomes, and they fall on the obligated provider or deployer rather than on their suppliers. It is on this page once because procurement teams ask, and nowhere else.
This page and briefing are educational material, not legal advice. Whether any Article 50 obligation applies to your organisation depends on your role, your systems and your use cases.
ECZ-ID provides identity and evidence infrastructure. Organisations remain responsible for determining their legal obligations.
MCP traffic between machines is not automatically Article 50(1) activity
Article 50(1) attaches to AI systems intended to interact directly with natural persons. An MCP connection between an agent and a tool server is machine-to-machine plumbing — it is not, by itself, a natural-person interaction, and operating an MCP server does not automatically place a provider in Article 50 scope.
The commercial pressure runs through a different channel: the enterprises deploying agents against your MCP surface are themselves accountable for what those agents do and disclose. They will increasingly ask who an agent is, who operates it, whose authority it carries and what transparency posture sits behind it — and they will prefer counterparties who can answer with portable, re-checkable evidence rather than a PDF attachment and a promise.
Assess my evidence — the £395 fixed-scope entry point
The £395 MCP Readiness Audit is a fixed-scope, human-reviewed assessment of one MCP implementation: protocol conformance, transport behaviour, authorisation boundaries, tool-surface risk and operational readiness.
Its existing scope already covers the ground most agent-accountability conversations turn on:
- Authorisation boundary review — how identity, scopes and tool-level permissions are, or are not, enforced.
- Tool-surface review — schema quality, input validation, destructive-action safeguards and prompt-injection exposure of tool descriptions.
- Operational readiness — logging, secret handling, deployment and rollback posture.
Article 50 or agent-accountability questions? Flag them in your written intake. We will confirm in writing whether they fit the agreed MCP assessment boundary before work begins.
No sales call required. TrustOps shows the complete scope and order before payment. After payment, complete the written intake; EcoCitizenz confirms scope and the delivery window before work begins.
Make your MCP more valuable to enterprise agent customers
Your gateway can authenticate a request. ECZ-ID can help answer who the agent is, who operates it, and what portable evidence it publishes.
Enterprises adopting agents are starting to ask their MCP providers questions the protocol alone cannot answer: which agent is calling, who operates it, whose authority it carries, and what transparency evidence stands behind it.
ECZ-ID is developing a portable evidence layer for exactly this: letting an MCP provider resolve supported, public-safe information about an agent and its operator before reliance, and request additional proof where the evidence supports it.
- Works alongside your existing authentication.
- Integrates with OAuth/OIDC and enterprise gateways.
- Adds portable agent and operator evidence for your existing policy layer.
- Agent identity
- Operator
- Principal
- Credential state
- Published regulatory evidence
- Evidence freshness
- Resolver reference
Available now as a design-partner engagement while the product interface is being standardised. Pilot partners shape the evidence fields, the resolution interface and the request-for-proof flow against their real enterprise customers' questions, and are first in line as the layer matures.
MCP provider? Start an Agent Evidence pilotBuild the evidence stack your use case actually needs
One credential rarely answers everything. Match the gap your customers, reviewers or counterparties are actually asking about — and start there.
THE FOUNDATION
ECZ-ID Business Passport™
The Parent Business Passport is your organisation's persistent ECZ-ID identity foundation and the authority spine every downstream credential resolves back to.
Already included where it is required: the MCP and agent plan prices above are all-in with the parent entitlement they need, and paid MCP services carry the applicable 90-day Parent benefit once payment is authenticated and provisioning succeeds. It is never charged twice, and payment alone does not verify an organisation.
Review & buy →Not every system needs every credential. Article 50 does not require these ECZ-ID products. Start with the evidence gap your customers, reviewers or counterparties actually need answered.
Ask which evidence fits your use caseThe executive briefing
Eight pages a CISO, legal lead or product owner can read in about six minutes — and act on the same day. Free, direct download. The optional email copy exists for your convenience, not as a gate.
8 pages · approximately 6-minute read · Primary-source references · Updated August 2026
Inside the briefing
- What Article 50 actually requires from 2 August 2026.
- The difference between provider and deployer obligations.
- Direct human AI interaction versus background and machine-to-machine operation.
- Machine-readable marking and disclosure requirements for synthetic content.
- What evidence enterprise buyers and auditors may ask to see.
- How portable, resolver-verifiable evidence can reduce repeated proof work.
Get the Article 50 briefing
Download PDF nowImmediate access. No phone number required.
PDF, 8 pages. No email required.
Or email yourself a copy
We use your information to provide this resource and measure its performance. Marketing updates are optional. See our Privacy Notice.
Free tool: scan your MCP configuration
The free MCP scanner analyses a pasted MCP configuration passively, in your browser: transport patterns, credential exposure, package launch hygiene and authorisation signals. Nothing is uploaded, nothing is executed, and live endpoint probing happens only in a separate, separately consented mode.
It examines MCP configuration hygiene — it does not assess Article 50 applicability or any legal obligation. It is the fastest way to see whether the machine surfaces behind your agents are ready for the questions this page raises.
Article 50, briefly answered
- Does Article 50 apply to my MCP server?
- Not automatically. Article 50(1) attaches to AI systems intended to interact directly with natural persons; machine-to-machine MCP traffic is not, by itself, that. Your role, your systems and your use cases determine which — if any — Article 50 paragraphs apply. This briefing is educational material, not legal advice.
- Do AI agents need a government credential now?
- No. Article 50 creates transparency obligations for specific roles and situations. It does not create an agent licensing scheme, and no government credential is required for an AI agent. Evidence infrastructure such as ECZ-ID is voluntary — useful where buyers ask for proof, never legally required.
- What changed on 2 August 2026?
- The Article 50 transparency obligations of the EU AI Act became applicable: interaction disclosure for providers, machine-readable marking of synthetic content, and deployer disclosure duties for emotion recognition, biometric categorisation, deep fakes and certain published AI text. Providers of generative systems already on the market before 2 August 2026 have a limited transition until 2 December 2026 for the Article 50(2) marking obligation.
- Does ECZ-ID make me Article 50 compliant?
- No. ECZ-ID helps organisations structure, publish and re-check identity, accountability and supporting evidence. It does not itself determine or certify legal compliance — your organisation, with its own advisers, decides what the law requires and whether you meet it.
Choose your next step
Start
Purchasable MCP and agent plans, published prices, operated through TrustOps. Free tier available.
See plans