Back to blog
B2B Prospecting

How to Add a Credit Approval Gate to AI Lead Search

A two-phase workflow for previewing contacts, setting a reveal cap, and requiring confirmation before an AI assistant uses lead credits.

RadarLeads·October 26, 2026·4 min read

If an AI assistant can search for contacts and reveal their details, separate those steps. Review the free, masked preview first, then approve a specific maximum number of new reveals. Radar Leads documents search_leads and get_credits as free, while reveal_leads uses one credit per newly revealed contact; contacts your organization already revealed are not charged again.¹

The prompt is only an instruction; the host client must control whether a reveal call can run without your confirmation.

Preview before authorizing a reveal

Start by telling the assistant what kind of contacts you need and the maximum number you might reveal. Ask it to use search_leads to show the filters, total matches and masked preview, then use get_credits to report the current balance. Both are free according to Radar Leads’ MCP documentation.¹

Review the search scope before moving on. A broad match count is not an instruction to reveal that many records. Tighten the filters or reduce the proposed batch if the preview is larger than the test you intended.

Set a worst-case credit budget

Choose a hard cap, N, before the reveal. Since each newly revealed contact costs one credit, N is the maximum possible charge for a request to reveal N contacts. The actual charge can be lower when some matching contacts have already been revealed by your organization.¹

The MCP tool accepts a requested count and has no documented built-in maximum. Larger reveals run asynchronously, so do not rely on the assistant to infer a sensible limit from a vague request like “get me some leads.” State the cap yourself and ask the assistant to stop if the balance or filters change before approval.

get_credits is a snapshot, not a reservation. If other people or automations use the same account, the available balance may change after the check. Recheck immediately before approving a reveal when that matters.

Require confirmation in the MCP client

MCP tools can be invoked by a model based on the conversation. The MCP specification recommends keeping a human able to deny tool calls and presenting confirmation prompts for operations.²

In your agent client, enable tool-call confirmation for reveal_leads if the client supports per-tool approvals. Do not treat a prompt as a technical restriction: if the host can execute the reveal without asking, the prompt alone cannot guarantee that it will pause. Check your client’s actual controls; if you cannot require confirmation for credit-consuming calls, do not rely on the prompt as a spending safeguard.

Copyable two-phase instruction

Use this as a starting point, then adapt the filters and cap to your own task:

For this Radar Leads search, do not reveal contacts yet.

First, use only the free search preview and current credit-balance tools. Show me:
- the exact filters you used
- the total matching count and a masked preview
- the maximum number of contacts you propose to reveal
- the worst-case credit cost for that count
- the current available credit balance

Then stop and wait. Do not call reveal_leads unless I send a separate message approving the exact filters and maximum count. If the filters, count, or balance changes before the reveal, show me the new details and ask again. Never exceed the approved count.

After an approved reveal, check its job status and report the requested and revealed counts. Do not export or start another credit-consuming action without separate approval.

Check the job before using the contacts

After you approve a reveal, Radar Leads returns a job ID. Use check_reveal_status to see progress; if the job is still processing, check again after a short wait. The tool documentation says contact details are returned inline for reveals of up to 100 leads; larger jobs can be exported after completion.¹

Compare the requested count with the completed count before using or exporting the results. If a job stops because the credit balance runs out, some contacts may already have been unlocked. A count cap and a host confirmation reduce the chance of an unexpectedly large reveal, but they do not replace checking the completed job.

For current tool names and limits, check the Radar Leads MCP guide before connecting an agent. If you want to try the service, create a free account.

Sources

  1. Radar Leads, “MCP Server & AI Agent Integration”
  2. Model Context Protocol, “Tools”

More from the blog

How to Create a B2B Buyer Persona from Customer InterviewsTurn past customer decisions into a cautious, evidence-led buyer persona with interview prompts, a note-taking framew...How to Find Decision-Makers at a Target Company Without Guessing AuthorityUse an account-first worksheet to identify plausible roles, record supporting evidence and separate a title from prov...How to Build a Targeted B2B Contact ListDefine your buyer criteria, review contact fit, and prepare a B2B list for outreach without confusing database matche...
Back to blog