MCP server (AI assistants)

secuveri runs a remote MCP server at https://mcp.secuveri.com/mcp, so AI assistants such as Claude, ChatGPT, Claude Code and Cursor can use your account without code: create verification links, read outcomes, build and publish workflows, manage block and allow lists, run AML screenings and search these docs. It uses the public API with the same rules, prices and limits.

Add it to your assistant

  • Claude and ChatGPT: add a custom connector with the URL https://mcp.secuveri.com/mcp.
  • Claude Code: claude mcp add --transport http secuveri https://mcp.secuveri.com/mcp.
  • Cursor and other MCP clients: add a server with the URL https://mcp.secuveri.com/mcp (streamable HTTP).
Cursor mcp.json
{
  "mcpServers": {
    "secuveri": { "url": "https://mcp.secuveri.com/mcp" }
  }
}

The first time, the assistant opens Connect with secuveri: you sign in, choose the organization and the environment (sandbox or production) and approve what the assistant may do. There are no API keys to paste. You can see the connection and disconnect it in the dashboard under Settings, Connected apps. Start with sandbox: no real people are involved and nothing is charged.

Tools

ToolWhat it doesNeeds
get_account_infoThe organization, environment, allowed scopes and go-live status-
create_verification_linkStarts a verification and returns the link for the personsessions:write
search_verifications, get_verificationOutcomes: status, verdict and reason, never personal dataresults:read
list_workflows, get_workflow, get_workflow_catalogYour workflows, and the steps, options and prices to build oneworkflows:read
create_workflowSaves a draft workflowworkflows:write
update_workflow, publish_workflowChanges or publishes a workflow; asks you first, as it changes live checksworkflows:write
screen_person_amlScreens a name against sanctions and PEP lists; charged per call, asks firstaml:screen
list_list_entries, add_list_entry, remove_list_entryBlock and allow listslists:read, lists:write
explain_reason_code, search_docs, get_code_sampleReason codes, these docs, and Node.js and Python samples-

The server also offers every guide and the OpenAPI description as resources, and three prompts: integrate_age_check, build_workflow and review_declined_verifications. Webhook management is not exposed.

Safety

  • Answers carry the same summaries as GET /v1/verifications: no documents, faces, names, birth dates, session tokens or signed results.
  • Tools that change something (publishing a workflow, a charged AML screening, removing a list entry) are marked so the assistant asks you before it calls them.
  • Text typed by your integration or by end users (vendor_data, metadata, list reasons) comes back marked as untrusted data, so the assistant does not follow instructions hidden in it.
  • The connection only gets the scopes you approved. Over the connection it sees its own verifications unless you allowed all of the organization's.