How to Enrich Contacts Inside Claude, ChatGPT, and Cursor (2026)

Until recently, using AI for contact enrichment meant copy-pasting between tabs: you'd ask Claude to draft an email, then leave the conversation, open your data tool, look up the prospect's email and phone, copy it back, and continue. The AI helped with the writing but couldn't touch your data.

The Model Context Protocol (MCP) changed that. It's an open standard that lets AI assistants connect directly to external tools and data sources. Connect an enrichment provider's MCP server to Claude, ChatGPT, or Cursor, and the AI can look up verified emails and mobile numbers inside the conversation — no tab switching, no copy-paste. You ask in plain English; the agent makes the call and returns the result.

This guide covers the mechanics that apply to any MCP-compatible enrichment tool, then walks through a concrete example. The setup pattern is nearly identical across providers — once you understand it for one, you understand it for all.

Last reviewed: 1 October 2026. TargetWise workspace MCP uses https://targetwise.ai/api/mcp with a bearer header. Host availability and authentication support vary; check current official client documentation.

1. What you need before you start

Three things, regardless of which AI client or data provider you use:

• A compatible MCP client. It must support remote HTTP MCP and custom Authorization headers for TargetWise’s workspace bearer endpoint. Native cloud connectors may require OAuth, which this endpoint does not currently implement.
• An enrichment provider with an MCP server. Several B2B data tools now expose one. You'll need an account and, in most cases, an API key from that provider's dashboard.
• The provider's MCP server URL and auth method. Usually a URL like https://mcp.provider.com/mcp plus either an API key (bearer token) or an OAuth sign-in.

2. How enrichment-over-MCP actually works

Worth understanding before you wire it up, because it explains what the AI can and can't do. MCP is a standard interface between an AI client and an external tool. When you connect an enrichment provider's MCP server, the provider exposes one or more "tools" — for example, an enrich_contact tool that takes a name and company and returns a verified email and mobile.

In a compatible authorised client, the model can call a discovered enrichment tool and present its actual response. Available fields and uncertainty must remain visible; the client does not turn a returned email into deliverability or an unknown business phone into a verified mobile.

This matters for two reasons. First, the AI handles the orchestration — you don't write code or format API calls. Second, the data quality is only as good as the provider behind the tool. MCP is the pipe; the provider is the water. Connecting a weak data source through MCP just gives you weak data faster. For the architecture behind why some MCP enrichment servers outperform others, see our companion piece on B2B email and mobile enrichment MCP servers.

3. How to enrich contacts in Claude

Claude remote connectors are account-brokered; local Claude Desktop MCP configuration is a separate mechanism. The current remote-connector guide includes Free, Pro, Max, Team and Enterprise users, with one custom connector for Free users. Team and Enterprise administrators control installation.

TargetWise’s workspace endpoint requires a custom bearer header and has no public OAuth. The documented Claude remote-connector setup does not establish custom-header compatibility, so do not promise a direct connection by pasting a key into the UI. Use an approved compatible local client or authenticated integration after verifying its current configuration and secret handling.

Current Claude remote connector documentation

4. How to enrich contacts in ChatGPT

ChatGPT’s authenticated MCP app guidance expects OAuth 2.1. TargetWise’s workspace bearer endpoint does not currently implement that flow, so these workspace instructions do not establish native ChatGPT connector compatibility. Do not paste an API key into a conversation or remove authentication to make a connector work.

An approved integration must satisfy the host’s authentication requirements while keeping workspace credentials on the appropriate server or client boundary. Confirm current plan and administrator permissions in the host; a qualifying plan alone does not fix authentication incompatibility.

Current ChatGPT MCP authentication guidance

5. How to enrich contacts in Cursor

Cursor documents remote MCP configuration with a URL and custom headers. In its MCP configuration, add the server below, supplying the key through the environment or your approved secret mechanism. Reload or reconnect, then inspect discovered tools and test a known record.

{
  "mcpServers": {
    "targetwise": {
      "url": "https://targetwise.ai/api/mcp",
      "headers": {"Authorization": "Bearer ${env:TARGETWISE_API_KEY}"}
    }
  }
}

Configuration is host-specific: VS Code uses a servers object and HTTP type; Claude Desktop local servers and other clients have their own transports and settings. Follow the documentation for the exact client and version.

Cursor MCP setup · VS Code configuration reference

6. Worked example: enriching with the Targetwise MCP

Here's the full flow end to end, using Targetwise as the example provider. The same pattern applies to any MCP enrichment tool — swap the URL and key for whichever you use.

1. Get a workspace key. Use your authorised workspace key and current subscription balance. Check the current pricing page for trial availability; no fixed 50-enrichment offer is established by this setup guide.
2. Add the MCP server. Use the URL YOUR_TARGETWISE_MCP_GATEWAY_URL with a bearer-token Authorization header — in Claude's Connectors UI, Claude Desktop's config, or Cursor's mcp.json.
3. Restart or reload the client. The enrich_contact tool appears in your tool list.
4. Prompt in plain English. Ask the AI to find an email and mobile for a named contact, or to enrich a list.

A typical exchange the agent handles natively:

For this workflow, use the current TargetWise plan rates and usage calculator. Paid plans renew monthly; unused balance resets at renewal.

7. Prompts that work (and ones that don't)

The quality of what you get out depends heavily on how you ask. The AI needs enough identifying information to match a contact, and a clear action.

Use one supported identifier combination. A LinkedIn URL alone is an alternative to a name plus one company identifier, not another field to add to it. Confirm identity first; extra unsupported fields can cause validation errors rather than improve matching.

8. Hypothetical MCP interactions, end to end

These examples explain a workflow; they are not transcripts, real people, measured results or verified contact details.

Single known-person lookup

Ask for available work email for a supported LinkedIn URL, or for Alex Example at example.com using first_name, last_name and company_domain. Request only needed fields and review the actual response.

A bounded list

Iterate one record at a time with an explicit maximum record count, budget and retry limit. Log unresolved fields and stop at the budget boundary. Do not infer a native batch endpoint from a conversational prompt.

Enrich, then act

Keep enrichment separate from sending or CRM writes. Review identity, mailbox status, phone type and permission for the intended action before the caller-controlled next step.

Refine an ambiguous record

Resolve the identity first, then submit one supported identifier combination. A response may include work_email or business_phone; returned emails do not prove deliverability and phones remain unclassified unless evidence identifies their type.

9. Troubleshooting: when the tool won't connect

Most setup failures come down to five causes. Work through them in order — the first two fix the large majority of cases.

Frequently asked questions