Skip to content
New · Model Context Protocol

Connect your AI agent.
Let it screen for you.

ScreeningHub ships an official Model Context Protocol server. Any MCP-aware agent — Claude, Cursor, Codex, your in-house orchestrator — can run PEP and sanctions checks, monitor profiles, and pull audit logs through your account, with the same scopes and audit trail as a human user.

Connect in 60 seconds See available tools

Compliance, on autopilot.

Screening is a perfect agent task: structured input, structured output, well-defined decisioning. Let your agent triage the inbox, your case manager catch up on the backlog, your customer-success bot clear KYC review queues — overnight, in parallel, with a full audit trail.

Triage the queue

"Clear every onboarding from the last 24 hours. Flag anything ambiguous." The agent runs the screens, summarises hits, and presents only the cases that need a human.

Investigate matches

When a customer hits a list, ask the agent: "Pull the underlying record, summarise the regulatory rationale, and draft an EDD memo." Reviewer keeps the final call.

Audit on demand

"Show me every screen we ran for customer X-12345 in the last 18 months." Agent fetches the audit trail, formats for the regulator, drops a CSV in your shared drive.

Connect in 60 seconds.

One config block in your MCP client. Auth via your existing API key.

claude_desktop_config.json
~/Library/Application Support/Claude/ 
{
  "mcpServers": {
    "screeninghub": {
      "command": "npx",
      "args": ["-y", "@screeninghub/mcp@latest"],
      "env": {
        "SCREENINGHUB_API_KEY": "sk_live_…"
      }
    }
  }
}
Claude — Desktop & Web
Cursor — settings → MCP
Any MCP client — stdio & SSE

Tools the agent can call.

Exposed under your existing API-key scopes. Read-only by default; --write opts in to mutating tools.

screen — Run a single PEP/sanctions/adverse-media screen

Args: name, optional type (person/company), country, date_of_birth, lists. Returns a pass/review/hit decision, matched records and an audit_id.

screen_batch— Screen many records in one call

Args: records[] up to 1,000 per call. Useful for "screen all customers onboarded yesterday" agent flows.

monitor_profile— Add or remove a profile from ongoing monitoring (write)

Args: customer_id, action (add / remove), identity_payload. Requires --write scope.

get_match— Fetch the full match record by ID

Returns the matched list record(s), source URLs, list version, full PEP role / sanctions designation history.

list_pending_reviews— Read open review-decision cases

Filters: since, severity, list_category. The agent's queue.

submit_review— Record an analyst decision (write)

Args: audit_id, decision (cleared / escalated / blocked), reasoning. The agent's action is logged with the agent's session identity.

export_audit— Pull a date-range audit log

Args: customer_id or since/until, format (json / csv). The deliverable for any regulator request.

Example agent flow

"Clear last night's queue and write the EDD memos."

User → Claude

"Run sanctions and PEP screens on every customer onboarded since 23:00 last night. For any review-flagged hit, pull the match record, summarise why they're on the list, and draft a one-paragraph EDD memo. Drop the open cases in #compliance-review."

Claude (using ScreeningHub MCP)

  1. Calls screen_batch with last night's onboardings.
  2. Filters review/hit from the response.
  3. Calls get_match for each open case → reads the source records and list rationale.
  4. Drafts the EDD memos (templated paragraph + list link).
  5. Posts a Slack thread with each open case & memo.
  6. Logs every action against the agent session ID for audit.

Built for regulated agent workflows.

Letting an agent take action on compliance data is exactly as serious as letting an analyst do it. The MCP server is built around that.

  • ✓Scoped API keys. Issue an agent-scoped key with read-only or read-write tools, separate from your service-to-service keys.
  • ✓Per-call audit trail. Every MCP-driven call is logged with the agent session ID and the tool name. The export looks identical to a human reviewer's audit row.
  • ✓Approval gates. Mutating tools (submit_review, monitor_profile) can be marked as "human approval required" — the agent must surface the action to the user before it commits.
  • ✓Rate limiting. Per-key call ceilings prevent runaway loops.
  • ✓EU data residency on Enterprise — the MCP server runs in the same region as your data plane.

Try it with the free tier

100 free screens per month, no credit card. Get an agent key, drop the config in your MCP client, run your first screen from a chat.

Get started

For background see sanctions screening explained and what is PEP screening.