Anthropic's MCP roster between October 2025 and May 2026 has pulled expert-network, ratings, and fundamentals vendors into a single callable surface inside Claude, with Guidepoint , Third Bridge, AlphaSense, Moody's, Hebbia, Daloopa, and Aiera all live as connectors.
The buy-side procurement model is shifting from per-call expert spend, per-seat market intelligence licences, and per-feed data contracts toward a unified context-consumption line tied to the LLM workspace.
Enterprise commit structures (visible in Hebbia's Claude Marketplace listing, where buyers can route existing Claude spend) let firms redirect committed AI dollars toward third-party connectors, which compresses traditional vendor pricing power.
Vendor positioning is following the budget: AlphaSense at USD 500M ARR , Daloopa's USD 47M Series C , and Rogo's USD 160M Series D all read as routing-layer pitches rather than seat businesses.
Compliance is the lag indicator. The Rogo bank-demo block reported in May 2026 shows MNPI and connector-permission review at large banks is still catching up to the procurement reality.
The pre-MCP procurement stack, line by line The shape of a buy-side research budget circa 2023 was familiar to anyone who had run a multi-strategy hedge fund or a mid-market PE platform. Expert-network spend sat in one bucket, usually per-call or per-seat, split across Guidepoint, Third Bridge, and GLG, with the healthcare pod or the consumer pod tracked separately so the head of research could push back on heavy users. Market intelligence sat in a second bucket, seat-based, dominated by AlphaSense at USD 500M ARR as of 2026 , with a long tail of Sentieo-style alternatives. Ratings sat in a third bucket, an enterprise license with Moody's or S&P that the credit pod used and the equity pod largely ignored. Fundamentals sat in a fourth bucket: FactSet, Bloomberg, and increasingly Daloopa for the parsed-filings layer, contracted as data feeds with their own seat counts, their own legal review, and their own renewal cycle.
Each of those buckets had a different owner inside the firm, a different procurement cycle, and a different unit of consumption. The expert-network bucket was reviewed against call volume; the market-intelligence bucket against seat utilisation; the ratings bucket against credit-pod headcount; the fundamentals bucket against feed uptime and coverage. There was no shared denominator. A CFO asking which vendor was producing the best return per dollar got four different answers in four different units.
The MCP roster collapses that structure. When the same Claude session can call Guidepoint transcripts, AlphaSense filings, Moody's ratings, and Daloopa fundamentals as context for a single research question, the unit of consumption stops being a call, a seat, a license, or a feed. It becomes a context request, billed through a wallet the buyer already controls.
What the MCP roster actually changes The technical shift is narrow. Model Context Protocol gives an LLM workspace a structured way to call out to a third-party library and pull defined slices of content back into the session, with provenance metadata attached. Anthropic's own framing of the Claude connector launch makes the developer story explicit: one client, many servers, predictable schemas. The commercial shift is broader, because the connector list now includes most of the libraries a buy-side analyst would otherwise touch through a separate vendor portal.
Guidepoint's exposure of its expert-call transcript archive as a Claude connector means an analyst building a thesis on, say, a US managed-care payer can pull relevant transcript excerpts into the same Claude session that is already reading the 10-Q. Third Bridge, similarly, exposes its Forum transcripts. AlphaSense exposes its filings and broker-research index. Moody's exposes ratings and rationale. Hebbia routes through its own document-reasoning layer. Daloopa exposes parsed historical fundamentals. Aiera exposes earnings call transcripts and intra-quarter monitoring. The set is not yet complete (FactSet and S&P Capital IQ remain conspicuous absentees as of mid-2026, and Bloomberg's terminal data is fenced for its own reasons), but the buyer's working assumption inside the LLM workspace is that the library will be there.
What that does to procurement is straightforward. The buyer no longer needs a separate contract for each library if the consumption pattern is itself becoming context-call dominated. Enterprise commit dollars sitting with Anthropic can be redirected toward third-party connectors via the marketplace structure visible in Hebbia's Claude listing. The procurement department's view of vendor spend collapses toward a single denominator: routed context volume, attributable per provider, billed against a wallet whose top-line number was negotiated once at the enterprise-commit level.
The unit-of-consumption shift, in one frame The most useful way we have found to describe this shift to research operations leads is as a unit change. The pre-MCP unit was the access right: a seat, a call pack, a feed. The post-MCP unit is the context call: a unit of routed library content, priced per token or per request, attributable to the provider whose library answered. A firm consuming USD 4M of research content per year through the old structure had four contracts to renew, four unit definitions to reconcile, and four utilisation reports to audit. The same firm consuming USD 4M through MCP has one wallet, one consumption log, and a vendor split that is computed from the routing data rather than negotiated up front.
The consequence for buyers is that vendor concentration becomes a runtime decision rather than a contract decision. The healthcare pod that needs more Guidepoint context this quarter can simply route more requests there; the credit pod that needs more Moody's context next quarter can do the same. There is no seat to redistribute, no call pack to true up, no feed to provision. The CFO sees one line.
The consequence for sellers is that the renewal conversation moves from seat counts to share of context. The vendor that wants to grow inside a buyer no longer pitches more seats; it pitches deeper library coverage, lower latency, and stronger provenance, because those are the dimensions that determine which connector the LLM calls when an analyst asks a question that could be answered from more than one library. That is a meaningfully different sales motion, and it favours vendors whose libraries are deep and whose retrieval is fast.
Vendor positioning has already moved The vendors themselves are not waiting for buy-side procurement to formalise this. AlphaSense's reported USD 7.5B funding round and its disclosure of USD 500M ARR read, in our view, as a routing-layer pitch rather than a market-intelligence-seat pitch. The dollar number that matters for an LLM-mediated workspace is not the number of seats sold but the share of context calls answered, and AlphaSense's investor narrative leans into that.
Rogo's USD 160M Series D , reported in May 2026, fits the same shape. Rogo's product surface is an investment-banking-flavoured LLM workspace that orchestrates calls to underlying libraries, which puts it in routing-layer territory by construction. Daloopa's USD 47M Series C is positioned around being the canonical fundamentals connector inside an LLM session, which is a meaningfully different pitch from selling parsed filings as a data feed. Aiera's consortium-backed sell-side content platform, similarly, is presented as the connector for earnings call content rather than as a transcript subscription.
The common thread is that the vendors with the cleanest narrative for the next eighteen months are the ones positioning their library as a connector inside whichever LLM workspace the buyer has standardised on. The vendors with the muddiest narrative are the ones still selling seats and call packs and trying to retrofit MCP exposure on top. We read the funding round sizes in this cohort as the market's bet on which positioning will hold.
What this does to the expert-network call pack The expert-network business model deserves a separate paragraph, because it is where the unit change is sharpest. A traditional expert-network contract priced calls. Guidepoint, Third Bridge, and GLG built their revenue around how many expert engagements a firm consumed in a quarter, with overage and underage tracked carefully. The MCP exposure of Guidepoint's transcript library does not eliminate the live-call business, but it changes the denominator. When an analyst can pull a relevant excerpt from a past Guidepoint expert call into a Claude session as context, without scheduling a new call, the marginal value of the live call shifts toward the cases the library cannot answer.
Our read is that this pushes expert-network revenue toward a barbell. One end is high-value bespoke live engagements, which the library cannot substitute for because the question is new or the expert is being asked to react to a specific scenario. The other end is library-as-connector revenue, billed against context calls, which compounds with the size of the underlying transcript archive. The middle, where a firm books a call to ask a question that is largely already answered by an existing transcript, compresses.
This is not a forecast that expert networks shrink. The deep libraries at Guidepoint, Third Bridge, and GLG are exactly the asset that makes the connector pitch credible, and the firms with the largest historical archives are best positioned to monetise routed context. The forecast is that the revenue mix changes shape, with library access becoming a more visible component of the contract and call packs becoming the variable layer on top.
Compliance is the lag indicator The procurement reset is running ahead of the compliance reset. The Rogo demo block at a large bank reported in May 2026 is the visible example: the buyer was ready, the vendor was ready, the workspace was ready, and the bank's compliance team paused the workflow on the perfectly reasonable grounds that the connector permission model had not yet been reviewed against MNPI policy. That is the right response from a compliance function, and it is a leading indicator of how the next twelve months play out at regulated buyers.
The specific questions compliance teams will need to answer are well-defined. Who in the firm is authorised to route a context call to which connector, and how is that audited. How is MNPI risk in expert transcripts re-screened when an LLM pulls excerpts into a synthesis. How does the firm prove provenance on a Claude-generated note that drew on six different vendor libraries. How does the connector-permission model interact with information barriers between research and trading. None of these are insoluble, and most have analogues in the way firms already manage seat-level access to expert networks. But they have to be solved before the procurement consolidation can be formalised in a regulated buyer's vendor list.
Our view is that compliance review at large banks lags the procurement model by roughly two to four quarters. Multi-strategy hedge funds and PE sponsors, whose compliance functions are smaller and more flexible, are already modelling connector spend as a unified line. Large banks will get there, but the demo blocks will continue in the interim, and vendors selling into bank buyers should expect a longer cycle on connector approval than on traditional seat or feed approval.
Three scenarios for the next eighteen months We think the next eighteen months break into three roughly MECE paths. In the base case , MCP-style connector procurement becomes the default at multi-strategy hedge funds and PE sponsors by the end of 2026, with large banks following on a one to two quarter lag once compliance frameworks settle. Connector spend appears as a unified line in the research budget, vendors compete on library depth, latency, and provenance, and the seat-and-call-pack contract structure becomes the variable layer rather than the base layer. In this world, the funding round sizes for AlphaSense, Rogo, Daloopa, and Aiera look correctly calibrated.
In the bull case for vendors , the connector model expands faster than the base case, the marketplace billing structure visible in Hebbia's Claude listing becomes the dominant procurement path, and a meaningful share of enterprise Claude commit dollars routes through to third-party libraries. In this world, the buyer's negotiating leverage on individual vendor renewals is high (because anything is a connector swap away), but the total addressable spend on routed context grows fast enough that vendors are net winners on revenue even if unit pricing compresses. The library-depth incumbents (Guidepoint, Third Bridge, AlphaSense, Moody's) benefit disproportionately.
In the bear case , compliance friction at regulated buyers, combined with model-vendor lock-in concerns, slows the procurement consolidation. Buyers continue to run parallel seat and feed contracts alongside connector access for at least another twelve to eighteen months, the unified context-spend line stays an internal management construct rather than a procurement reality, and vendor pricing power on traditional units holds up longer than the bull case implies. Even in this scenario, our view is that the direction of travel is the same and only the timing slips.
What we would ask an expert next For a research operations lead or head of procurement at a buy-side firm thinking through this, the questions that surface most usefully in our conversations are the practical ones. How does the firm attribute routed context spend back to the requesting pod when the LLM pulled from four libraries to answer one question. What is the firm's policy on connector-vendor concentration: is there a target share-of-context for any single library, and how is it monitored. How does the firm value the historical transcript archive of an expert network when most of the consumption is now library access rather than new calls. How does the firm's MNPI policy treat an LLM synthesis that draws on multiple vendor libraries. What is the firm's view on routing through a model vendor's marketplace versus contracting connectors directly. The answers vary by firm, but the firms with crisp answers are the ones that have moved fastest on the procurement reset.
Why it matters The durable shift is that the buy-side research budget is collapsing toward consumption-based AI context spend, with connector providers competing on library depth, latency, and provenance inside one buyer wallet. The vendors that win the next eighteen months are the ones whose libraries are deep enough to be the default connector for a given question type and whose retrieval is fast enough to stay default at runtime. The vendors that struggle are the ones whose pricing depends on seats or call packs that the LLM workspace has quietly made variable. Our read is that the procurement consolidation is already underway at the most flexible buyers, that compliance review at regulated buyers will lag by two to four quarters, and that the funding round sizes in this cohort are the market pricing in a unit-of-consumption change rather than a temporary AI cycle.
