The Licensed-Access MCP Layer: Publishers and Database Owners Sell Agent Entry to Their Corpora — With Published Price Points

The Licensed-Access MCP Layer: Publishers and Database Owners Sell Agent Entry to Their Corpora — With Published Price Points

A librarian-curated comparison maintained by HKUST Library (updated Sep 2026) documents — in one place — the layer of the research-tool market where data owners expose curated corpora to external AI agents via MCP, with explicit pricing and access rules¹. This is the academic sibling of the financial-data MCP unbundling (Institutional MCP Financial Data Servers: FactSet, S&P Global, and PitchBook Unbundle) and the supply side that makes the destination-vs-backend bifurcation work (Destination vs. Backend: The Research-Tool Market Bifurcates — and Both Layers Converge on MCP).

The price points

Wiley Scholar Gateway MCP — "First major academic publishers to launch an official MCP server for AI agent access," covering "8M+ articles from more than 2,000 Wiley journals, searching full text rather than abstracts only." The pricing is the first published consumer-facing price for publisher full-text via MCP: "Free-tier limits (2026-09) – Limit the free plan to 20 results per query and 30 queries per month; the paid plan is 500 queries per month (US$15/month)." It returns RAG chunks (not whole articles), preserves "citation metadata, DOI links, peer-review status, and method context," and is expanding to partner publishers (Sage, ASM).

Clarivate Nexus Connect (Apr 28, 2026) — an "institutional MCP gateway that routes ChatGPT or Claude queries to a university's library collections and Clarivate content," i.e. a university-branded connector inside ChatGPT and Claude that can also "check print availability and place holds without leaving the conversation." Its companion Nexus Extend is a browser extension doing "real-time citation verification" against trusted indexes — AI applied after retrieval, mapping LLM output back to institutional holdings.

EBSCOhost AI Exchange / Bridge MCP (May 2026) — connects "licensed EBSCOhost full text to external AI platforms through an MCP connector," including a Perplexity Computer partnership. Separately priced beyond the bundled database AI.

Plus a long tail of MCP endpoints on the same pattern: PubMed (full-text limited to open-access PMC), Scite (papers + 90M+ patent families + clinical trials + grants in one connection), Consensus, Elicit (Pro+ only), Undermind, Statista, O'Reilly.

The publisher workspace counterpoint: LeapSpace

The same page documents what Elsevier built on its own corpus instead of renting it out: LeapSpace, "Elsevier's next-generation 'research-grade AI workspace' launched in Jan 2026... unify[ing] the capabilities of Scopus AI and ScienceDirect AI into a single secure platform which supports the full researcher workflow, from literature discovery and funding identification to collaborator discovery and hypothesis generation." Data: "16M+ peer-reviewed full-text articles and books from Elsevier and licensed subscription and open access articles from other leading publishers... along with 105M+ abstracts from Scopus." Its verification feature is the telling one: "Trust Cards – Show the exact passage used to support each AI-generated insight for easy verification." Limitation: "Extra subscription cost on top of Scopus."

What this means for the market map

• The data-owning incumbents split into two strategies: rent the corpus (Wiley, EBSCO, Clarivate — MCP endpoints priced per query or per institution) vs. build the workspace (Elsevier's LeapSpace — the full-workflow destination on proprietary full text). Clarivate is doing both (Nexus Connect gateway + Nexus Extend bridge).
• Full text is the moat, abstracts are the commodity: every tool built on Semantic Scholar/OpenAlex inherits the post-2024 abstract restrictions from Elsevier and SpringerNature — which is exactly why the publishers who own full text can charge for MCP access.
• Verification UX is converging across the layer (Trust Cards, citation-integrity preservation in Wiley's gateway, Nexus Extend's citation checking) — evidence-passage linkage is becoming table stakes, not differentiation (AI Is Breaking the Proxies for Expertise: Goodhart's Law Comes for Research Output).