MSCI

Limits

• MSCI's own terms name MCP connectors as a prohibited extraction route. MSCI's published Notice and Disclaimer states that, without prior written permission, the Information must not be used "to train, use as an input to, or otherwise in connection with" any artificial intelligence, machine learning or large language models, except as licensed and expressly authorized under MSCI's AI Contracting Supplement — and separately bars accessing or extracting the Information using "unauthorized bots, scrapers, crawlers, AI agents, MCP servers or connectors". MSCI's connector guide says content generated through the connector is subject to your own MSCI licence agreement and points you to your licence terms. The connector is presumably the authorised route; the published terms do not say so in as many words. Get your position confirmed in writing by your account manager. We are describing a documented tension, not giving legal advice.
• Sustainability and climate ratings are opinions, in MSCI's own words. MSCI states that its "sustainability and climate offerings, research and data are produced by, and ratings are solely the opinion of MSCI Solutions", and that such materials "have not been submitted to, nor received approval from any regulatory body". Its controversies research is described as allegations from vetted public sources rather than assertions of fact. An ESG Rating returned through this connector is an opinion an agent will render as a value.
• MSCI disclaims all warranties of accuracy and completeness. Its notice states, in capitals, that no Information Provider makes any express or implied warranties or representations with respect to the Information or the results obtained by using it, and expressly disclaims implied warranties of "ORIGINALITY, ACCURACY, TIMELINESS, SUITABILITY, NON-INFRINGEMENT, COMPLETENESS, MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE". The user assumes the entire risk of any use made of the Information.
• The Information is explicitly not investment advice, and not a substitute for your own judgment. MSCI states it "does not constitute, and must not be relied upon as, investment advice, credit ratings, or proxy advisory or voting services", that no Information Provider is a fiduciary or makes any recommendation or endorsement of an investment decision or asset allocation, and that the Information "should not be relied on and is not a substitute for the skill, judgment and experience of any user". All Information is described as impersonal and not tailored to any person.
• Past performance, back-tests and signals carry named caveats. MSCI states the Information, including index construction, ratings, historical data or analysis, is not a prediction or guarantee of future performance. Back-tested performance is hypothetical, not actual, and frequently differs materially from results later achieved. Signals are described as "inherently backward-looking" and "inherently inaccurate", not intended to predict the future.
• Private asset valuations are estimates. MSCI states private capital, equity, credit and other assets may be assessed infrequently, may not be priced on a secondary market, and must not be relied on as a valuation of a particular instrument; that reliance on fair value estimates and non-market inputs introduces bias and subjectivity; and that IRR metrics "are not fully representative" without full disclosure of fund cash flows, assumptions and time horizons. That last one bears directly on the Private Capital Intel tools.
• Redistribution is barred by default. MSCI's notice states that without prior written permission you must not redistribute or resell the Information in any form, permit a third party to access or resell it, include it in materials for public dissemination such as fund factsheets, market presentations or prospectuses, use it to create or as a component of any financial product, use it to create or optimise indexes, ratings or other data products, or use it to populate a database. Your own licence may vary this; the published notice does not.
• MSCI discloses a conflict of interest on rated issuers. Its notice states that issuers mentioned in MSCI Solutions materials, or their affiliates, may purchase research, products or services from MSCI affiliates, or manage funds rated by MSCI Solutions or based on MSCI indexes, and that MSCI Solutions has taken steps to mitigate potential conflicts.
• MSCI states it may use AI to generate the Information itself. Its notice says MSCI and its Information Providers "may use automated technologies and artificial intelligence to help generate content and output incorporated in the Information".
• No real-time data. MSCI states index data is updated generally on a T+1 basis with no intraday data; Private Capital Intel benchmarks update quarterly; Real Assets Index data follows each market's publication cycle, typically quarterly or annual; and TPM reflects whatever has been ingested to your Total Plan Manager account.
• History is capped. MSCI documents up to 20 years of history for standard indexes, subject to data availability and your licence, and as little as 12 months for custom indexes.
• Named index modules are not reachable through the connector at all. MSCI lists Float, Thematic Exposure, Business Segment, Constituent AUM, Investability Data, Sustainability Index Methodology Data, Factor Index Methodology Data and the Index Rebalance Criteria Report among modules not yet available through IndexAI Insights, along with custom indexes delivered solely via vendors.
• Complex queries can time out or return incomplete results. MSCI documents that long-running queries "may time out or return incomplete/incorrect results" from processing or context-window constraints, and advises breaking requests up. An incorrect result from a truncated query is a documented outcome, not a hypothetical one.
• Custom index construction and non-standard periods are out of scope. MSCI states that queries needing custom index construction, custom segmentations or background processing are unavailable through the connector, as are annualised figures for non-standard horizons such as two or seven years.
• Some figures are calculated by the model, not by MSCI. For Real Assets, MSCI states any figure computed by the AI rather than returned by the MSCI API will be labelled "LLM-calculated — not MSCI-generated". We could not verify that this labelling extends to the other product lines.
• Entitlement gating is silent. Queries outside your licence return no results rather than an entitlement error the client can distinguish, and separate licences apply within product lines — Infrastructure and Real Estate Debt datasets are named as needing their own.
• We could not verify tool behaviour ourselves. The endpoint returns 401 to an anonymous handshake, so no input schemas and no safety annotations were readable. No tool on this server carries a safety annotation we can see. The read-shaped classification of the 24 get/list/ search/read/query tools rests on naming and documentation, not on annotations.
• The two non-read tools are unresolved. Neither execute_index_gql_schema nor run_at_workflow_step is documented by MSCI. We cannot say whether the GraphQL executor accepts mutations, or what a workflow step does or whether it is reversible.
• The vendor documentation URL in the directory is behind a login wall. It resolves to a Liferay portal login that immediately posts a SAML request to MSCI's identity provider, so it is unreadable without MSCI credentials. This page is thin-sourced by necessity on anything only that article covers. Everything above comes from MSCI's public connector user guide, its public product pages and its public legal notice instead.
• Ten of twenty-six tools have no vendor documentation we could find — the seven at_pai tools and the three sustainability and climate tools. Their groupings on this page are inferred from naming, not read from MSCI documentation.
• Two abbreviation expansions are inferred, not confirmed. at and pa are our reading of MSCI's product naming. pci, tpm and rai are confirmed by MSCI's user guide.
• The prompt is unconfirmed. Anthropic's snapshot lists one prompt name, Welcome_text, and the gated endpoint blocked our own check, so we cannot say what it contains.
• This page describes tools. It gives no investment or financial advice, and takes no position on how any figure, rating or benchmark returned by this connector should be used.

Frequently asked questions

Sources

• MSCI Connector user guide, May 2026 edition (PDF) — https://www.msci.com/downloads/web/msci-com/data-and-analytics/private-asset-solutions/ai-for-private-markets/msci-mcp-msci-connector-user-guide.pdf (fetched 2026-08-19; the primary vendor source for this page and publicly reachable without login, though marked for authorized MSCI clients. Source of the four capabilities, the PCI/TPM/Real Assets Index expansions, prerequisites, the web-search-off instruction, entitlement behaviour, update cadences, history caps, known limitations and the notice and disclaimer. Search engines still index its title as naming three integrated MCP tools; the edition fetched names four) · retrieved 2026-08-19
• MSCI Notice and Disclaimer — https://www.msci.com/notice-and-disclaimer (fetched 2026-08-19; source of the AI and MCP-connector extraction restriction, the AI Contracting Supplement requirement, the redistribution and derivative-works prohibitions, the warranty disclaimer, the not-investment-advice statement, the ratings-are-opinion statement, the back-test and Signals caveats, the private-asset valuation caveats and the issuer conflict-of-interest disclosure) · retrieved 2026-08-19
• MSCI Supplemental Terms for Client Use of Artificial Intelligence — https://www.msci.com/legal/supplemental-terms-for-client-use-of-artificial-intelligence (fetched 2026-08-19; the licence route the Notice points at for permitted AI use. The operative terms are in linked PDFs — an AI Contract Supplement, an Approved AI List and a client letter — which we did not review. The landing page states clients must contact MSCI for AI use beyond its scope) · retrieved 2026-08-19
• MSCI Private Assets AI Connector — https://www.msci.com/data-and-analytics/private-asset-solutions/ai-for-private-markets/private-assets-ai-connector (fetched 2026-08-19; MSCI's marketing page for the connector, naming Private Capital Intel, Total Plan Manager, Index Intel and IndexAI Insights, OAuth-authenticated access and entitlement-controlled data. No endpoint URL and no read/write statement) · retrieved 2026-08-19
• MSCI Private Capital Indexes and Private Capital Intel product pages — https://www.msci.com/data-and-analytics/private-asset-solutions/msci-private-i/private-capital-intel (fetched 2026-08-19; corroborates the pci expansion) · retrieved 2026-08-19
• MSCI Total Plan Manager product page — https://www.msci.com/data-and-analytics/private-asset-solutions/total-plan-manager (fetched 2026-08-19; corroborates the tpm expansion and the transparency-analysis surface) · retrieved 2026-08-19
• MSCI AI Portfolio Insights product page — https://www.msci.com/our-solutions/analytics/risk-management/ai-portfolio-insights (fetched 2026-08-19; the basis for reading at_pai as AI Portfolio Insights. This is our inference, not an MSCI statement about the tool prefix) · retrieved 2026-08-19
• Live anonymous probe — 2026-08-19: an anonymous initialize call to the endpoint returns HTTP 401 with a JSON INVALID_TOKEN body and a www-authenticate header naming realm="mcp-gw" and an authorization_uri pointing at the path-append protected-resource descriptor. We did not authenticate and called no tool · retrieved 2026-08-19
• Live RFC 9728 protected-resource descriptor — all three path forms checked 2026-08-19. Root and path-insert return HTTP 404 with an identical JSON error body; path-append (/.well-known/oauth-protected-resource/mcp/v1.0/mcp) returns HTTP 200 with a descriptor naming one scope, offline_access, plus RS256 signing, header bearer methods and https://mcp.msci.com as the authorization server · retrieved 2026-08-19
• Live RFC 8414 authorization-server metadata — https://mcp.msci.com/.well-known/oauth-authorization-server (fetched 2026-08-19; issuer https://mcp.msci.com, flow endpoints on accounts.msci.com, PKCE S256, three client-auth methods, a gateway-hosted registration_endpoint, and 14 scopes of which only offline_access overlaps with the resource descriptor). /.well-known/openid-configuration returns 404 on the same host · retrieved 2026-08-19
• Anthropic connector directory — https://claude.ai/directory/2cc10749-99cb-4853-a9d7-b2292ec589fc (snapshot 2026-08-16; tool_names lists 26, prompt_names lists one — Welcome_text — permissions reads "Read and write", and the record carries a works_with list of Claude, the Claude API and Claude Desktop. Added to the directory 2026-02-16) · retrieved 2026-08-16
• MSCI connector documentation URL from the directory — https://support.msci.com/support/shared-content/msci-mcp-connector-for-claude/06779542517 (checked 2026-08-19 with a browser user-agent; redirects to a Liferay portal login at /c/portal/login which auto-posts a SAML authentication request to accounts.msci.com. Unreadable without MSCI credentials. No claim on this page rests on it) · retrieved 2026-08-19
• robots.txt on www.msci.com — https://www.msci.com/robots.txt (fetched 2026-08-19; disallows a long list of application and asset paths. No Content-Signal axes are declared. support.msci.com/robots.txt returns an empty Disallow:, permitting everything, though the content behind it is login-gated anyway) · retrieved 2026-08-19
• llms.txt on www.msci.com — https://www.msci.com/llms.txt (checked 2026-08-19; returns HTTP 404 with a Next.js error shell. MSCI publishes no llms.txt) · retrieved 2026-08-19
• MSCI Privacy Notice — https://www.msci.com/legal/privacy-notice (the directory's privacy_policy URL; listed for completeness)