Sumsub MCP server icon

Sumsub

by Sumsub

HIPAA CompliantSOC2 ReadyISO 27001 Ready
Finance18 tools

Read KYC/AML verification data and create applicants from an AI agent. Anthropic lists 18 tools; three write, and none approves or rejects a verification. OAuth sign-in behind a single opaque mcp scope.

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 Sumsub via MCP

https://api.sumsub.com/mcp/

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

Sumsub Tools & Capabilities (18)

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

Limits

  • The OAuth scope is a single wildcard. mcp names no object and no verb. Nothing at the grant separates reading an AML case from creating an applicant, and a future tool added to the server falls inside the same consent already given.
  • Sumsub publishes no tool list. The 18 names come from Anthropic's directory. Sumsub's MCP page names tools only inside four examples, and one of those — applicant_moderation_state_list — is not in Anthropic's listing at all.
  • We could not read tool schemas or safety annotations. The endpoint returned 401 to an anonymous request, so no parameter-level detail is published here, including for how_to_sumsub, whose behaviour we could not characterise.
  • No accounting tool exists. Verifications bill per applicant at a published $1.35–$1.85, and no listed tool reports balance, usage or cost. Billing surfaces a day late.
  • Whether a generated link is delivered to the applicant is unresolved. The underlying endpoint accepts an email and phone number. We found no documentation stating whether Sumsub sends a message, and we did not test it. Treat it as unknown rather than as a no.
  • MCP is not a documented audit-trail source. The audit trail costs extra, and its documented xClientId values are API, Dashboard and SDK. You may not be able to separate agent actions from human ones.
  • Fraud-network results are inferences, not findings. Labels such as "Potential link to fraudulent applicant" can rest on a similar device. Do not let an agent restate them as established fact about a person.
  • Rate limits are tight on AML reads. Sumsub's default API limits are 300 GET and 50 POST requests per 5 seconds, but its AML case endpoint documents a much lower 10 requests per minute — a batch triage across a queue will hit it.
  • Network-based fraud links require an entitlement. Sumsub states these are visible only if Fraud Networks is enabled for your client key, and that enabling it goes through a Customer Success Manager, so applicant_fraud_network_list may return nothing on an account without it.

Frequently asked questions

No tool among the 18 Anthropic lists carries an approve, reject, decline or reset verb. Sumsub's platform does support manual approval and rejection, but through the Dashboard and API endpoints that have no counterpart in this listing. The connector can read a verification verdict and create new applicants; it cannot change an existing applicant's status.

Identity document extractions, AML screening hits and fraud-network links about named people. Sumsub's applicant data response includes idDocs fields carrying first, middle and last names, issuing authority, issue and expiry dates, plus nfcInfo, beneficiaries and questionnaire answers. This is among the most sensitive personal data any connector in the directory reaches.

Not through any listed tool. The 18 names include no image, document-download or file-fetch verb, and applicant_get maps to Sumsub's applicant data endpoint, which returns structured fields extracted from documents rather than the scans themselves. Sumsub's separate document-image endpoints have no counterpart in this listing.

Two of the three write tools can start a billable check. Sumsub's published pricing is 1.35 to 1.85 US dollars per verification depending on plan, and its billing documentation states checks appear on the Billing page the following day, not the day they run. No tool in the listing reports balance, usage or cost.

Sumsub operates an audit trail API, but it is sold at additional cost and its documentation does not name MCP as a source. The event fields include an xClientId describing the source of an action, documented with the values API, Dashboard or SDK. Whether MCP calls appear, and under which value, is not documented.

One, and it names no object and no verb. The server's RFC 9728 descriptor advertised scopes_supported of exactly mcp on 2026-08-23. A single unqualified scope cannot separate reading an applicant from creating one, so consent is all-or-nothing across the whole tool surface.

A Dashboard role permission called Use MCP server, plus the role's existing permissions. Sumsub documents that the agent performs the actions allowed by your Sumsub role permissions, so the connector inherits rather than widens your access. The permission sits in the Manage group alongside managing dashboard users and application tokens.

We could not establish that it does, and the parameter is worth knowing about. verification_link_create maps to Sumsub's external WebSDK link endpoint, whose request body accepts applicantIdentifiers containing an email address and phone number. Sumsub's MCP documentation describes generating a link, not delivering it, and does not say whether supplying those identifiers triggers a message.

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

Server Info

Category
Finance
Developer
Sumsub
Tools
18
Domain
api.sumsub.com

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