Data access fees lose to open platforms

The first shows how it starts. On October 1, it announced an expanded “flexible ecosystem”: API access, data warehouse connections, a sandbox, a certification program, and MCP access for integration providers coming later. The release mentioned no fee. Six days later, vendors were telling trade press what it would cost them to stay connected.

The second is Envestnet. Citywire reported on November 2 that Envestnet is weighing MCP access fees for integration partners. MCP is a standard way AI tools connect to software.

What the terms say, and what vendors say it costs

The platform’s published terms describe a monthly ecosystem support fee charged to technology providers. It is billed monthly and in advance, it can change on 30 days’ notice, and access can be suspended or limited for non-payment. That is the documented part.

In trade press reporting in October, more than six vendors quoted fees between $2,000 and $20,000 a month, with the “vast majority” at $20,000. One small vendor put its bill at $120,000 a year. The platform cited cybersecurity and AI as reasons for the program and said clients own their data. By the same reporting, it left 7 of 13 questions unanswered, including whether a vendor authorized by a client can reach that client’s data without paying. One anonymous vendor called it “reprehensible to charge for MCP access.”

Further coverage added mechanics:

The fee is a flat monthly charge per vendor, tied to consumption. First invoices go out in late October and are due in November, and usage-based pricing follows in the first quarter of 2027. Haik Sahakyan of ARQA said in trade press coverage, “An ecosystem can technically remain open while becoming economically difficult.” Michael Roth, also quoted in that coverage, expects large vendors to fold the cost into renewals, which means the RIA pays, and smaller vendors to drop the integration.

That is the part that lands on your desk. A software vendor you rely on is deciding whether to absorb a monthly bill that most of the quoted vendors put at $20,000, pass it to you, or stop maintaining the connection. None of those choices involves your client, and none of them is yours.

Banks and software vendors are building meters too

This is not only a wealthtech story. Banking got there first. CNBC reported in November 2025 that JPMorgan had struck paid data-access deals with Plaid, Yodlee, Morningstar, and Akoya. The Financial Technology Association called them “prohibitive tolls” and “anti-competitive.” PNC and Wells Fargo are following.

The CFPB’s 2024 rule under section 1033 banned such fees. A federal court in Kentucky enjoined that rule on October 29, 2025 while the CFPB reconsiders it, and a new proposal has sat at OIRA since August 6, 2026. It may allow “reasonable fees.”

Enterprise software is moving the same way. SAP began routing third-party AI agents through metered pathways in June 2026, and Forrester called that “the wrong call.” Salesforce will meter every registered-agent call over MCP or API in Flex Credits, with the rate still to be announced. Microsoft charges per Dynamics tool call for agents outside Copilot Studio. Atlassian will meter enriched cross-product calls at $0.01 per credit from December 3, 2026, while basic reads stay free.

US antitrust law rarely forces anyone to open up. Under Trinko, US antitrust law seldom compels a company to deal with a rival, and a fee is not illegal per se. The risk rises with dominance and pretext. The EU has gone further: its Data Act bans cloud switching and egress charges starting January 12, 2027. I have seen no regulator say anything about wealthtech data fees.

That last sentence matters for timing. Where nothing regulates a toll, it can last for years. Banking is the live example: JPMorgan is already charging aggregators, and the one rule that banned those fees is enjoined. Wealthtech has its own version in switching costs. Moving a reporting or portfolio platform is painful enough that many firms will absorb a passed-through fee and stay. Some of the fees announced now will stick for a long while. They give way to open platforms over time.

Included MCP, no extra toll

A longer list of companies includes MCP access in the platform itself. Datadog made its MCP server generally available, free, on March 9, 2026. On the May 7 earnings call, its CEO reported MCP calls up four times quarter over quarter while human web use also grew. Google Cloud offers MCP “at no extra cost” to paying customers, and GitHub offers it to all users regardless of plan. Notion and Linear include it in every plan, free ones too. HubSpot made MCP generally available to all accounts, read and write.

Linear’s Cristina Cordova wrote on October 6 that weekly MCP use had grown roughly twenty times in a year, from a smaller base, while weekly app users also grew and paid seat utilization is higher than two years ago.

The Datadog and Linear numbers are their own reporting, and they show that a company can open access to agents and keep growing.

Who charges to connect, and who doesn’t, as of October 17, 2026.

Health care already wrote the rule

Health care is where US law has already written an answer down. The 21st Century Cures Act, implemented through ONC rules, defines information blocking as a practice by a provider, a certified health IT developer, or a health information network that is likely to interfere with access, exchange, or use of electronic health information. Fees have their own exception. Those actors may charge businesses a cost-based fee, even with a reasonable profit margin, but not one based on whether the requester is a competitor or on the value the requester derives. For patients the rule goes further: no fee is allowed that is based in any part on electronic access by the individual, their representative, “or another person or entity designated by the individual.”

On December 19, 2025, ONC published FAQs saying that interfering with access by “automation technologies (e.g., robotic process automation (RPA), agentic artificial intelligence)” could be information blocking, because the rule is not specific to any means of access. The penalties are large: up to $1 million per violation for certified developers and networks. Enforcement is still early, and as of August 1, 2026, no OIG penalty had been publicly reported. I have found none since.

HIPAA still allows a reasonable, cost-based fee; for an electronic copy, a provider may charge a flat fee of up to $6.50 or its actual or average allowable cost. The point that matters here is this: under the information blocking rules, patient-directed electronic access carries no access fee. On the payer side, health plans fall under HIPAA rather than the Cures exclusion, and HIPAA limits what they can charge an individual to reasonable, cost-based copying costs.

Agents are already acting for patients. OpenAI picked b.well on January 7, 2026 to connect users’ medical records to ChatGPT Health, through FHIR APIs and trusted exchange networks with identity verification and consent.

Financial data deserves the same deal

Financial data should not be treated significantly differently from health data. If a patient, and the apps and agents the patient authorizes, can get electronic access to health records without access tolls, then a client, and the tools their advisor uses on their behalf, should get financial data the same way.

Here is how I build that from the rules themselves. Health law gives a right plus a fee ban for patient-directed electronic access. Finance has the statute but not a working right. Section 1033 of Dodd-Frank says a covered provider “shall make available to a consumer, upon request” information about the product it provided, but “subject to rules prescribed by the Bureau.” The CFPB’s 2024 rule covered checking-type accounts, credit cards, and payment facilitation. It never covered brokerage or advisory accounts. It is enjoined, and the CFPB has no enforcement authority over SEC-regulated firms, which includes registered investment advisers.

The privacy statutes do not fill the gap. GLBA and the regulations that implement it, the CFPB’s Regulation P and the SEC’s Regulation S-P, require privacy notices, opt-outs on sharing, and safeguards. None of them gives a customer the right to get a copy of their own data. HIPAA does, with fee limits and an enforcement record of 56 OCR right-of-access actions.

On fee logic, health rules bar any fee based on electronic access by the patient or anyone the patient designates, and they bar fees tied to a business requester being a competitor or to the value it derives. Financial data access has no such rule today, so fees like the ones now being announced are set privately.

I haven’t found anyone making this comparison in print. The health model is the better template for clients, because it bars fees for access by the person the data is about and by the tools that person designates.

The strongest arguments for the fee

Security

The strongest is security. Amended Regulation S-P requires firms to oversee their service providers, and an AI agent reaching into a system of record is a new kind of access. But look at what oversight requires: knowing who has access, authenticating them, logging what they touch, and being able to cut them off. Those are reasons to govern access. None is a reason to price it. Suspending access for non-payment is a billing control, and FTC staff have said they will read security rationales for blocking interoperability closely.

Cost

The second is cost. Agents can call an API far more often than a person clicking through screens, and serving that traffic is not free. Fair. But serving agent traffic is part of running the platform, and it belongs in what customers already pay for it, as Google Cloud and GitHub show. I have seen nothing connecting the quoted figures to a cost basis, and if the anonymous quotes are right that most vendors land at the same $20,000, that sits awkwardly with a measurement of load.

The moat

The third is the moat. Muboli argued in January 2026 that lasting pricing power sits with whoever controls permissions, approvals, and audit. That is right, and it is the best argument against a connection fee. The moat shifts to who may see what, who approved the action, and an audit trail a compliance officer can read. Firms will pay for that inside the product. A fee on the mere act of connecting is a different thing, and it is the part a competitor can route around.

Where this ends

Data access fees lose to open platforms over time. Charging for data access harms the people who use the platforms that do it, and over time it harms the platforms too.

Start with who pays. If Michael Roth is right, the fee lands in renewals, so the advisor pays more for the same software or loses an integration a smaller vendor cannot afford to keep. Neither helps the client.

Then look at where the work starts. The AI interface becomes the place advisors begin their work: asking, drafting, reconciling, checking. The apps do not disappear. They sit behind it as systems of record, the place the data lives and the audit trail is kept. Once that is how people work, a platform’s value is what the agent finds when it arrives. Andreessen Horowitz wrote in March 2025 that agents choose tools on speed, cost, and relevance, and a toll fails the cost test first. Jason Lemkin of SaaStr made the practical version: metering pushes customers to copy their data out, which he called “a weekend project now,” and he said he will not adopt tools that tax his agents. He won’t be the last buyer to say so.

In the short run, the market will fragment. Traditional players charge, some vendors pay and pass it through, some drop integrations, and where nothing regulates the fee it may hold for years. In the long run, the platforms that include agent access in what customers already pay and compete on permissions, approvals, and audit win, because that traffic is where their customers’ work is going.