Ten thousand MCP servers are live, and your attribution vendor probably just announced one. The MCP project reports over 97 million monthly SDK downloads and 10,000 active servers. MCP based identity resolution in AI agents sounds like progress. It can also quietly corrupt your creator ROI numbers.
What is MCP doing in your attribution stack?
MCP (Model Context Protocol) is a standard that lets an AI agent call tools and data sources. It does not resolve identity. Your vendor’s matching logic does that, and MCP is the pipe the agent uses to reach it.
That distinction matters in a demo. A vendor says, “Our agent resolves identities through MCP.” Ask which part is the protocol and which part is proprietary. The protocol handles connection and permissions. The match logic, the graph behind it and the confidence scoring are all vendor-specific, and they decide whether a creator’s affiliate click ends up credited to the right customer.
Where do agents change the risk profile? They act without a human reviewing each query. An analyst who sees a suspiciously high match rate pauses and asks questions. An agent stitches the record and moves on to budget recommendations.
The match-rate trap
Every vendor leads with match rate. It is the easiest number to inflate and the least useful one on its own.
A coalition study from CIMM, run by Sequent Partners, Pre-Meditated Media and Janus Strategy & Insights, makes the point well. The CIMM report on match rate and bias says no match is perfect and that assessing match accuracy is essential to address measurement bias. It focuses on television data, but the lesson carries over to creator attribution. High and precise match rates together improve reliability. High alone just means the vendor matched a lot of things, some wrongly.
A match rate without a precision rate is a vanity metric. Ask for both, measured against a holdout you control.
Creator campaigns make this worse. Traffic from a TikTok or Instagram link often arrives through in-app browsers, with stripped referrers and no logged-in state. Matching leans on hashed emails, device signals or probabilistic inference. Our breakdown of identity graphs versus hashed email covers where each fails for creator traffic.
Five questions to ask before the pilot
- What is the match methodology, per identifier type? Deterministic (login, email) and probabilistic (device, IP) matches should be reported separately. If the vendor blends them into one figure, push back.
- How do you measure precision? Ask for a method, not a number. Acceptable answers include a known-ID validation set or a holdout test. Unacceptable: “our model is highly accurate.”
- What does the agent do when confidence is low? The safest behavior is to return an unresolved flag. The riskiest is to pick the best guess and keep going.
- Can I see the resolution trail? Every merged identity should be traceable to the evidence that produced it, with timestamps.
- Where does the data live? Know whether identity data stays in your warehouse or gets copied into the vendor’s graph.
Question three deserves a live test. Feed the agent a deliberately ambiguous record, such as two customers sharing a household device. Watch what happens. You will learn more in ten minutes than in a ninety-slide deck.
Permissions: where the protocol helps and where it doesn’t
The MCP authorization specification is clearer than most vendors’ security pages. The official authorization spec requires clients to implement Resource Indicators (RFC 8707), so a token is requested for a specific MCP server. It also warns about the confused deputy problem, where an intermediary server lets a downstream API trust a token it should not. Token passthrough, forwarding client tokens unchanged to downstream services, is explicitly forbidden.
Why should a marketer care? Picture an identity agent with access to your CRM, ad platforms and creator payout data. If its server forwards tokens carelessly, a request meant for one system can reach another. The result is a privacy incident and, almost as bad, contaminated data in your attribution model.
Ask the vendor three things:
- Do you validate token audience on every request?
- Do you pass tokens downstream, or issue separate credentials per service?
- Can I scope the agent to read-only access on identity tables?
A vendor who cannot answer these in plain language has not done the work. The deeper permission checklist is in our guide to MCP servers and your CRM.
Does the identity layer match your attribution foundation?
Here is a mistake I see often. A team buys an agent layer first, then discovers their identity foundation is a patchwork. The agent can only reconcile what the underlying systems expose.
Decide upfront whether you are relying on a customer data platform, a third-party identity graph or a platform’s own ID. CDP versus identity graph tradeoffs shape what any agent can resolve. Enterprise suites add their own wrinkles, which we cover in Microsoft, Adobe and Oracle identity risk.
Then ask what the vendor’s agent does with conflicts. If your CDP says a customer is one person and the vendor’s graph says two, who wins? A good vendor has a documented precedence rule, and you should be able to change it.
Run a test that exposes inflated attribution
Vendor-supplied case studies are marketing. Build your own test.
Step one: pick a past creator campaign with known outcomes, ideally one that used unique discount codes or tracked links. Those codes are your ground truth.
Step two: have the vendor’s agent attribute conversions without seeing the codes. Compare.
Step three: check false merges. How many conversions were credited to creators whose codes were never used? Over-attribution is the costly error because it moves budget toward the wrong partners.
Step four: run the same test with a holdout, a segment of audience not exposed to the creator content, to see whether the lift the agent claims is incremental. Our piece on incremental ROI claims explains how to design that holdout.
Ask for the test to be repeatable. If the agent returns different identity decisions on the same inputs a week apart, you have a stability problem. Some variation is expected with model-driven systems. Unexplained variation is not.
Contract terms that protect attribution accuracy
Protocol standards will not protect you after signature. Contracts will.
- Audit rights: the ability to review resolution logs and sample matched records.
- Accuracy reporting: quarterly precision and match-rate reports with methodology attached.
- Data deletion: proof that identity data is removed from vendor graphs on exit.
- Change notification: advance notice when matching models or connected data sources change, because a silent model update can shift your reported creator ROI overnight.
- Human override: the right to lock an agent’s actions to recommendation-only mode.
Change notification gets skipped most often. Yet a model swap mid-quarter makes trend lines meaningless. You would be comparing before and after on two different measurement systems.
Red flags worth walking away from
- A single headline match rate with no precision figure.
- “Proprietary” as the answer to every methodology question.
- No way to export resolution logs.
- Agents granted broad write access by default.
- No documented behavior for low-confidence matches.
One red flag is a conversation. Three is a no.
Next step: before your next vendor call, pick one past creator campaign with tracked codes and make a blind attribution test on that campaign a condition of the pilot. If the vendor will not run it, you have your answer.
FAQs
Does MCP resolve identities on its own?
No. MCP is a protocol for connecting AI agents to tools and data. Identity resolution depends on the vendor’s matching logic, the data sources behind it and the confidence rules the agent follows.
What is a good match rate for creator attribution?
There is no universal benchmark. A high match rate means little unless the vendor also reports precision, tested against ground truth such as discount codes or a holdout. Evaluate both together, as the CIMM study on match rates and bias emphasizes.
What is the confused deputy problem in MCP?
It happens when an MCP server acting as an intermediary lets a downstream API trust a token it should not. The MCP authorization spec explicitly forbids token passthrough to reduce this risk. For marketers, it means an agent could touch systems it was never meant to reach.
Should the agent have write access to identity data?
Start with read-only access and recommendation mode. Grant write permissions only after the vendor’s resolution logs have proven reliable over a pilot period, and scope them narrowly.
How do I test a vendor’s attribution accuracy?
Use a past campaign with unique codes or tracked links as ground truth. Have the agent attribute conversions blind, then count false merges and over-attributed creators. Add a holdout group to check incrementality.
Who should be involved in vetting?
Bring marketing analytics, legal or privacy and security teams to the same table. Marketers judge the accuracy, security validates token handling and legal covers data processing and deletion terms.
Top Influencer Marketing Agencies
The leading agencies shaping influencer marketing in 2026
Agencies ranked by campaign performance, client diversity, platform expertise, proven ROI, industry recognition, and client satisfaction. Assessed through verified case studies, reviews, and industry consultations.
Moburst
-
2
The Shelf
Boutique Beauty & Lifestyle Influencer AgencyA data-driven boutique agency specializing exclusively in beauty, wellness, and lifestyle influencer campaigns on Instagram and TikTok. Best for brands already focused on the beauty/personal care space that need curated, aesthetic-driven content.Clients: Pepsi, The Honest Company, Hims, Elf Cosmetics, Pure LeafVisit The Shelf → -
3
Audiencly
Niche Gaming & Esports Influencer AgencyA specialized agency focused exclusively on gaming and esports creators on YouTube, Twitch, and TikTok. Ideal if your campaign is 100% gaming-focused — from game launches to hardware and esports events.Clients: Epic Games, NordVPN, Ubisoft, Wargaming, Tencent GamesVisit Audiencly → -
4
Viral Nation
Global Influencer Marketing & Talent AgencyA dual talent management and marketing agency with proprietary brand safety tools and a global creator network spanning nano-influencers to celebrities across all major platforms.Clients: Meta, Activision Blizzard, Energizer, Aston Martin, WalmartVisit Viral Nation → -
5
The Influencer Marketing Factory
TikTok, Instagram & YouTube CampaignsA full-service agency with strong TikTok expertise, offering end-to-end campaign management from influencer discovery through performance reporting with a focus on platform-native content.Clients: Google, Snapchat, Universal Music, Bumble, YelpVisit TIMF → -
6
NeoReach
Enterprise Analytics & Influencer CampaignsAn enterprise-focused agency combining managed campaigns with a powerful self-service data platform for influencer search, audience analytics, and attribution modeling.Clients: Amazon, Airbnb, Netflix, Honda, The New York TimesVisit NeoReach → -
7
Ubiquitous
Creator-First Marketing PlatformA tech-driven platform combining self-service tools with managed campaign options, emphasizing speed and scalability for brands managing multiple influencer relationships.Clients: Lyft, Disney, Target, American Eagle, NetflixVisit Ubiquitous → -
8
Obviously
Scalable Enterprise Influencer CampaignsA tech-enabled agency built for high-volume campaigns, coordinating hundreds of creators simultaneously with end-to-end logistics, content rights management, and product seeding.Clients: Google, Ulta Beauty, Converse, AmazonVisit Obviously →