S&P Global — Deterministic Retrieval MCP server icon

S&P Global — Deterministic Retrieval

by S&P Global — Deterministic Retrieval

HIPAA CompliantSOC2 ReadyISO 27001 Ready
Finance37 tools

Query S&P Capital IQ financials, prices, transcripts and M&A data from an AI agent. Read-only: 37 tools in Kensho's source, 22 in Anthropic's listing, because the list is filtered by what your firm licenses.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect S&P Global — Deterministic Retrieval via MCP

https://kfinance.kensho.com/integrations/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

S&P Global — Deterministic Retrieval Tools & Capabilities (37)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • Your tool list is not this list. Thirty-three of the 37 tools require a dataset entitlement, so a firm that does not license transcripts sees no transcript tool. Plan against the tools your account actually returns, not against any published count.
  • No accounting tool exists, and no rate is published. Nothing in the surface reports usage, quota, credits or a balance — a scan of all 37 names for credit, balance, usage, quota and billing returns nothing. Kensho publishes no rate limit or per-call price for this server either. The sibling Adaptive Retrieval server, by contrast, documents quota, billed overage and usage headers, which shows Kensho meters where it means to. The cost of a Deterministic Retrieval session is defined by your contract and is not observable from inside the agent, so find out what your agreement says before running an agent in a loop against it.
  • The data is licensed, and the code licence does not cover it. The kfinance client is Apache 2.0 with a real LICENSE file, but the Capital IQ, Visible Alpha and transcript content it returns is separately licensed. Kensho's permission model includes an entitlement whose name explicitly excludes redistribution, so redistribution rights differ *within* a single subscription.
  • We could not read S&P Global's terms of use. www.spglobal.com returned HTTP 403 to every client we tried, including a browser user agent, so we could not check for clauses governing AI use, agentic workflows or reuse of retrieved output. Your firm's own S&P Global agreement and any unpublished addendum govern, and we have not read them. Do not treat this page's silence as permission.
  • Broad screening is not supported. Kensho states the API is company-object oriented and that a prompt must name at least one identifier, and that for broad industry research other S&P Global solutions may be better suited. The screening routes that exist in the REST API are not exposed as MCP tools.
  • Accuracy is your side of the line. Kensho's FAQ states that no AI model is used or provided as part of the service and that for LLM integrations accuracy depends on the model and architecture you use. The API itself is deterministic; the summary the agent writes from it is not. Kensho notes the generated code can be audited to verify which retrieval steps were taken.
  • One tool is staff-gated. get_issuer_ratings_from_identifiers requires a Kensho staff entitlement in the version we read, so credit ratings are not part of a normal subscriber's surface despite appearing in the source.
  • Two names in Anthropic's listing no longer exist. They were removed in Kensho version 7.0.0. A client that hard-codes them against the current server will fail.
  • This is not the Adaptive Retrieval connector. S&P Global lists two MCP servers with similar descriptions. Everything here describes Deterministic Retrieval at kfinance.kensho.com; Adaptive Retrieval is a separate endpoint with separate metering.
  • The nominated documentation URL is a signpost, not a document. Kensho's MCP overview page carries roughly 150 words of its own prose — the rest of the 344 words a text extractor sees is the site navigation. It is not a thin site, though: the substance sits on the sibling pages that navigation links to, and the full OpenAPI specification is served inside a __NEXT_DATA__ payload on the API page rather than as a fetchable spec file. All of them are listed below.

Frequently asked questions

Thirty-seven exist in Kensho's source; you will see fewer. Anthropic's directory lists 22. Kensho's kfinance package, which builds both the local and hosted servers, registers 37 and then filters them against the permissions on your account before the list is served. Thirty-three of those 37 tools are gated behind a dataset entitlement, so your firm's subscription decides the list.

No. All 37 tools in Kensho's source begin with get_ and every one resolves to a fixed read path on the LLM-ready API. The published OpenAPI specification describes 63 paths with 56 GET and 8 POST operations and no PUT, PATCH or DELETE anywhere. Each POST is a batch query envelope carrying identifiers, not a write.

Kensho publishes no per-call rate, quota or rate limit for this server, and no tool reports usage. Its sibling Adaptive Retrieval server is explicitly metered, with billed overage and quota headers on every response. Deterministic Retrieval has no equivalent published accounting, so treat spend as contract-defined and unobservable from inside the agent.

Named executives and directors, including their pay. The get_professionals_from_identifiers tool returns a roster with person_id, name, title and tenure, and an optional include_compensation flag adds reported pay. Its per-person companion adds include_biography, returning career narrative, prior employers and typically education. Both default to false, so the agent must ask for them.

Because Kensho filters the list by entitlement. Its client builds every tool, then keeps only those whose required permission intersects your account's permissions, so a firm without a transcripts licence never sees the transcript tool. Kensho states the same thing in its documentation: users can ask these questions depending on which datasets they are licensed for.

Read your S&P Global agreement first. The kfinance client is Apache 2.0, but that licence covers the code, not the data it returns. Capital IQ financials, transcripts and Visible Alpha estimates are separately licensed content, and Kensho's own permission model names one entitlement that excludes redistribution. We could not retrieve S&P Global's terms of use to check further.

Yes, and it is a separate listing. S&P Global — Adaptive Retrieval is a different server at grounding.kensho.com that takes a natural-language query and routes it across datasets itself. Deterministic Retrieval, described on this page, requires the agent to pick the tool and name a company. Anthropic lists both, and they can be used together.

A paid LLM-ready API subscription and Kensho Okta credentials. Kensho documents that access is arranged through S&P Global Marketplace and that your Okta credentials arrive by email when you sign up for a trial or account. Kensho also states you must be a Claude team owner to add the connector to your organisation for the first time.

Sources

Use in Agentman

Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio

Resources

Server Info

Category
Finance
Developer
S&P Global — Deterministic Retrieval
Tools
37
Domain
kfinance.kensho.com

Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.

Similar MCP Servers for Finance

Stripe logo

Stripe

by Stripe

Coming Soon
Finance23 tools
View detailsConnect
PayPal logo

PayPal

by PayPal

Finance28 tools
View detailsConnect
Square logo

Square

by Square, Inc.

Coming Soon
Finance3 tools
View detailsConnect

Related Resources

Browse All MCP ServersClaude & ChatGPT Setup GuideBrowse AI Skills