What Is an MCP Server for Identity Verification?

AI agents are moving from answering questions to taking actions: opening accounts, onboarding customers, moving money, screening counterparties. The moment an agent touches a real user or a real transaction, someone has to prove that the human is real and the money is clean. That is what an MCP (Model Context Protocol) server for identity verification does: it gives an AI agent a safe, structured way to run identity and fraud checks instead of guessing. This post explains what MCP is, why agents need identity and fraud tooling, and how Didit's hosted identity MCP server works end to end.

Key takeaways

• MCP (Model Context Protocol) is an open standard that lets AI agents call external tools through a single, consistent interface.
• An identity verification MCP server exposes KYC (Know Your Customer), KYB (Know Your Business), and fraud checks as callable tools an agent can invoke in a workflow.
• Didit's MCP server is the official hosted server at https://mcp.didit.me/mcp, with 130+ tools across 11 categories.
• Authentication is OAuth 2.1 + PKCE ("Log in with Didit") — no API key for the hosted server, and every action respects your console role.
• Pricing stays the same as the API: a full KYC core flow is $0.33, wallet screening $0.15/check, with 500 free verifications/month. The MCP layer itself is free.
• You can connect from Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed, and ChatGPT.

What MCP actually is

MCP (Model Context Protocol) is an open standard for connecting AI applications to external systems. Instead of every model vendor inventing its own plugin format, MCP defines one protocol: a server publishes a set of tools (each with a name, a description, and a typed input schema), and any MCP-compatible client — Claude, Cursor, ChatGPT, and others — can discover those tools and call them on the model's behalf.

The practical effect is that an AI agent stops improvising. Rather than pattern-matching text that looks like a verification result, the agent calls a real tool, that tool hits a real API, and a real, structured result comes back. For anything regulated — identity, compliance, payments — that difference is the whole point. You want an auditable action, not a hallucination.

Didit's server speaks MCP over Streamable HTTP at https://mcp.didit.me/mcp. You can use the hosted server (recommended) or self-host from the open-source repository.

Why AI agents need identity and fraud checks

Any agent that acts on behalf of a business inherits that business's compliance obligations. If an agent onboards a user, that flow still needs KYC. If it onboards a company, it still needs KYB and UBO (Ultimate Beneficial Owner) discovery. If it moves value, it still needs AML (Anti-Money Laundering) screening and transaction monitoring.

Without a dedicated tool, an agent has two bad options: fabricate a verdict, or bolt together brittle custom code for every check. An identity and fraud MCP server removes both problems. The agent asks for the check it needs — "verify this person," "screen this wallet," "monitor this transaction" — and the server returns a real result it can reason over and act on. The human stays in the loop for anything consequential, because the server is built to require confirmation on writes.

That is why the checks matter across the full lifecycle: authenticate (is the user real and present), verify (one-time KYC, KYB, biometric, document, AML screening), and monitor (transaction monitoring and ongoing screening after onboarding).

What a hosted identity MCP server does

A hosted identity MCP server is the bridge between an AI client and an identity-and-fraud platform. Concretely, Didit's server lets an agent:

• Create and read verification sessions for ID Verification, document OCR, and biometric liveness.
• Run KYB: registry lookup, officers, UBO extraction, and linked KYC for each beneficial owner.
• Screen individuals and entities against sanctions, PEP (Politically Exposed Person), and adverse-media sources.
• Screen crypto wallets for sanctions and mixer exposure, and monitor transactions in real time.
• Manage workflows, questionnaires, lists (blocklist / allowlist), cases, reports, alerts, and webhooks.

Because everything is exposed as typed tools, the agent composes them the way a developer would compose API calls — but in natural language. A realistic prompt looks like this:

"Create a KYC verification session for our standard onboarding workflow and give me the verification link to send to the applicant."

The agent picks the right tool, fills the schema, calls Didit, and hands back the link.

How Didit's MCP server works: OAuth, tools, and roles

Authentication uses OAuth 2.1 + PKCE — the "Log in with Didit" flow through business.didit.me. There is no API key to paste into a config file for the hosted server. When you first connect, your MCP client opens a browser, you approve access, and the server issues user-scoped tokens under two scopes: didit:management and didit:verification.

The important consequence is that the MCP server respects your console role. If you sign in as a Reader, the agent can list and read but not mutate. Compliance Officer, Developer, Admin, and Owner each unlock the actions their role already permits in the console. You do not get a second, looser permission model just because an AI is driving — the agent can only do what you can do.

Pricing is unchanged from the standard platform: a full KYC core flow is $0.33, wallet screening is $0.15/check, AML Screening is $0.20, transaction monitoring is $0.02/transaction, and every account gets 500 free verifications/month with sub-2-second inference. The MCP layer that exposes all of this is free.

Connecting your AI client

Setup depends on your client, but the pattern is the same: point it at the hosted URL and complete the OAuth login.

For Claude Code, add the server from your terminal:

claude mcp add --transport http didit https://mcp.didit.me/mcp

Then run /mcp inside Claude Code to trigger the "Log in with Didit" flow.

For Claude Desktop, go to Settings → Connectors → Add custom connector and paste https://mcp.didit.me/mcp. Cursor, VS Code, Windsurf, and Zed each accept an HTTP MCP entry in their JSON config pointing at the same URL. For ChatGPT, you can add the server as a connector by URL through OpenAI's Developer Mode — note that this is an OpenAI beta, so behavior may change.

The server is open-source (MIT) at github.com/didit-protocol/mcp if you want to read the implementation or self-host. There is no npm package to install — the path is the hosted URL plus OAuth.

Start building

An identity MCP server is what lets an AI agent do compliant work instead of pretending to. Didit runs it as free infrastructure on top of the same platform trusted by 1,500+ companies in production, across 220+ countries, 14,000+ document types, and 48+ languages — from a Y Combinator W26, profitable company that has raised $7.5M.

Read the full developer overview at didit.me/developers/mcp and the integration docs at docs.didit.me/integration/mcp/overview. When you are ready, Start free at business.didit.me — every account includes 500 free verifications each month, and connecting the MCP server costs nothing.