nvNeuralVerge

Building a Sales Intelligence Agent

An ai agent data api for sales: find and verify emails and phones, read LinkedIn and Crunchbase, over REST or MCP, so one agent can chain every step.

By the NeuralVerge team · 12 min read · Updated October 2026

▸On this page

A sales agent needs an ai agent data api that does more than return one record: it has to resolve who a person is, find a way to reach them and learn enough about their company to make the message worth sending. NeuralVerge covers all three from one catalog, with the same request format, over REST or MCP. This post walks through the five-step sequence, the source behind each step, and what it costs.

The short version

  • —The answer: chain five calls: a LinkedIn match on name and domain, a profile read, the email finder, email validation and a Crunchbase lookup. Each answers one narrow question, and each returns the same response envelope, so the agent needs no per-source parsing.
  • —Who it is for: teams building outbound enrichment, CRM backfills, lead scoring or account research, whether the caller is a script or an agent choosing its next step mid-task.
  • —What it costs: the sources in this post run from 1 to 150 points per call, at $0.001 per point. The full sequence for one person is at most 46 points, about $0.046, and each lookup in the sequence except validation is charged only when it resolves. Figures come from the source pages at the time of writing.
  • —How it connects: every source is a REST endpoint and a tool on the hosted MCP server. The source catalog lists each one with its exact fields and request shape.

The workflow in five steps

The order matters. Each step uses what the previous one produced, and the agent stops early when it already holds what it needs.

1. Resolve the person from a name and an employer

Most intake records, whether from a form, a CRM row or a list, hold a name and a company and nothing else. LinkedIn profile by name and domain takes a full name and a company name or domain and resolves the matching profile. It is usually the first call in the pipeline because a name and an employer is what the data you already have actually contains.

2. Read the profile, and the email if one is exposed

LinkedIn people profile plus email takes a profile URL and returns the person's name, headline, location and about text, plus an email in the same call when one is available. A profile does not reliably expose an address, so the agent must be ready for that field to be empty.

3. Find the address you still lack

When step 2 comes back without an email, the email finder takes a first name, last name and company domain and resolves a work address, with a confidence rating and a plain found flag to branch on. This is the call for a CRM record that has a name and an employer but an empty email field.

4. Validate before you send

Email validation takes an address and returns deliverability signals for 1 point, the cheapest call in the catalog. Run it on anything the finder returns, and on any address in a list you hold that never passed through a finder. A confidence rating tells you how well an address is corroborated; validation tells you whether the mailbox will accept mail today. They are different questions, and conflating them is a common source of bounced mail.

5. Add company context

Crunchbase company takes a Crunchbase company URL and returns total funding in US dollars, funding rounds, investors, founders, founding date, employee range and industries. A newly funded company is one of the more reliable buying triggers in B2B, and funding is the field no other step provides.

A worked example: one person at Acme Oy

Take an intake record with nothing but "Jane Example, Acme Oy, Finland." The agent runs the sequence above. The profile match resolves the name and acme.example to a profile URL. The profile read returns headline, location and about text, with no email this time. The finder takes the first name, last name and domain and resolves a work address. Validation checks it before it reaches a sending domain. Crunchbase resolves Acme Oy's funding history, so the brief can note whether the account raised recently.

Five calls, each answering one narrow question. The agent did not need to know in advance which source would succeed. It ran the sequence and recorded what came back, with each field tied to the call and session id that produced it. The same key also covers corporate registry records from dedicated registry sources and cited AI research for questions a record cannot answer, so the agent can extend the brief without a new integration.

What each source returns

Every one of these sources answers a different question, and knowing which question you are asking decides which call you make. None of them replaces another.

Email: two directions

The finder and enrichment run in opposite directions. The finder goes from a name and a domain to an address. Email enrichment goes from an address to the person: it returns full name, gender, phones, locations, company, position, work experience, and LinkedIn, X and Telegram handles where available. Use the finder for discovery and enrichment for identification.

Phone: worldwide or US-specific

Phone enrichment takes a phone number and returns the same kind of profile as email enrichment: name, company, position and social handles. It works worldwide. Phone enrichment US is a specialised version for US numbers, returning line type, carrier, a prepaid flag, an activity score and a litigator-risk flag. Reach for the worldwide source by default, and the US one when that line-level detail matters for a given workflow.

LinkedIn: five lookups, five starting points

Identity resolution has more than one entry point, so LinkedIn has five dedicated lookups. Pick the one that matches what you already hold.

LookupYou start withPoints
LinkedIn profile by name and domainA full name and a company name or domain10
LinkedIn people profile plus emailA profile URL10
LinkedIn people searchFilters: title, seniority, function, company150 per block of 25
LinkedIn company searchA query, industry, size or location5 per company
LinkedIn company employeeOne or more named target companies40 per run, plus 5 per profile

Points are from the source pages, at the time of writing. People search is the widest filter set in the catalog, and narrow filters keep the result relevant. Company employee suits prospecting scoped to a target account list rather than one company at a time. Company search returns industry, headquarters, follower count and a short description for each match.

Crunchbase: the funding signal

Crunchbase company costs 15 points per profile that resolves. Pair it with LinkedIn company search and you have both how the company presents itself and what it has raised. The figure is only as current as the Crunchbase profile itself, and undisclosed rounds do not appear.

One response shape across all of it

Every source answers in the same envelope: a session id, a kind naming the source with an echo of the request parameters, a readable human summary, a structured machine object and total_points. An agent that has learned this shape from the email finder already knows how to read a Crunchbase or LinkedIn response. That consistency is the point of one catalog instead of five vendor integrations with five formats. Each source is read at the moment the agent calls, not from a stored copy.

REST and MCP

The same request is reachable over both. Over REST, the finder looks like this:

curl -X POST https://api.neuralverge.ai/functions/v1/run-email-finder \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "first_name": "Jane", "last_name": "Example", "domain": "acme.example" }'

The response has the shape the source page documents, shown here with placeholder values:

{
  "session_id": "11111111-1111-1111-1111-111111111111",
  "kind": "email_finder",
  "domain": "acme.example",
  "first_name": "Jane",
  "last_name": "Example",
  "email": "jane@acme.example",
  "human": "# Email finder ...",
  "machine": {
    "found": true,
    "email": "jane@acme.example",
    "confidence": "high",
    "message": null
  },
  "total_points": 10
}

Branch on found rather than testing whether email is empty, and read confidence as a corroboration rating, not a delivery guarantee. The machine object also echoes the inputs and names the upstream provider; both are trimmed here.

NeuralVerge hosts an MCP server at https://mcp.neuralverge.ai/mcp, and every source in this post is a tool on it: run_email_finder, run_email_enrichment, run_email_validation, run_phone_enrichment, run_phone_enrichment_us, run_linkedin_domain (profile by name and domain), run_linkedin_email (profile plus email), run_linkedin_people_search, run_linkedin_company_search, run_linkedin_company_employee and run_crunchbase_company. REST paths use hyphens; the MCP tool names use underscores. The same finder call over MCP:

{
  "tool": "run_email_finder",
  "arguments": {
    "first_name": "Jane",
    "last_name": "Example",
    "domain": "acme.example"
  }
}

The response has the same envelope, at the same price. There is no separate MCP rate.

Connecting an MCP client

Add the URL to an MCP client such as Claude, ChatGPT, Cursor, VS Code or Gemini CLI, sign in through the browser and click Allow. There is no key to paste into a config file and nothing to host. The setup guide for Claude walks through it, and the client guides cover the others. Connected clients can be revoked under Settings, then API, in the dashboard, where they carry a "Connector:" label.

When to pick which

Pick REST when your code already knows when a call should happen, such as a nightly CRM backfill. Pick MCP when an agent has to decide mid-task which source to try next, for example moving from the finder to validation, or back to LinkedIn when the finder comes back empty. AI research is the one capability that runs over REST only.

Cost and limits

Each call is priced in points, and a point is $0.001. For the Acme Oy example, the profile match is 10 points, the profile read 10, the finder 10, validation 1 and Crunchbase 15, so 46 points, or about $0.046, in the case where every step resolves. A step that finds nothing is not charged, except validation, which is a flat 1 point per check. People search is the expensive end of the range, at 150 points per block of 25 results, so apply narrow filters before you page through results.

Two limits are worth stating plainly. A found email is a candidate, not a guarantee, and the finder's results depend heavily on the company: very small companies often have no discoverable pattern, and generic mailbox providers such as free webmail are not company domains and will not resolve a specific person. Point costs and fields come from the source pages and can change, so check the current page before you budget a large run.

Pitfalls

  • —Skipping validation. Sending to a found address without checking it is how a good list damages a sending domain.
  • —Starting at the wrong lookup. Calling people search when you hold a profile URL costs more and returns less than the lookup built for that input.
  • —Treating an empty field as an error. A profile with no email is a normal outcome. Let the agent route to the finder instead of retrying.
  • —Reading old funding as current. Check the date on the latest round before using funding as a trigger.
  • —Storing personal data without a reason. Email, phone and profile data is personal data. Keep what the workflow needs and decide the retention period up front.

How to decide

  • —Which starting point do you actually have? A name and a domain, a profile URL, or just a company. Match it to the lookup built for that input instead of the one at the top of your mind.
  • —Is the email field empty or filled? Empty means the finder; filled and unidentified means enrichment.
  • —Do you validate before you send? A found or enriched email is a candidate. Run it through validation before it reaches a sending domain you care about.
  • —Is a phone number worth the extra call? Use the worldwide source by default and the US one only when line type, carrier or litigator risk changes what you do.
  • —Do you need funding or only identity? Add Crunchbase when account fit or timing depends on what the company has raised.
  • —Should the sequence be fixed or decided by an agent? Fixed jobs such as outbound list enrichment, CRM backfills and lead scoring suit REST. Account research before a call, where the next lookup depends on what came back, suits MCP.

Frequently asked questions

Should I use the email finder or email enrichment?

It depends on what you start with. The email finder takes a first name, last name and company domain and resolves an address you don't have yet. Email enrichment takes an address you already have and returns the profile behind it. If the email field is empty, use the finder. If it's filled in and you want to know who it belongs to, use enrichment.

Do I need to validate an email after the finder returns one?

Yes, before any large send. A found address carries a confidence rating, not a deliverability guarantee. Email validation checks whether the mailbox is likely to accept mail right now, costs 1 point, and is worth doing before the address reaches a sending domain you care about.

Which LinkedIn endpoint do I call when I only have a name and a company?

LinkedIn profile by name and domain. It takes a full name and a company name or domain and resolves the matching profile, which is the usual starting point when a form or a CRM record gives you a person's name and employer but no direct link to their profile.

Can I get someone's email directly from their LinkedIn profile?

LinkedIn people profile plus email takes a profile URL and returns the email when one is available, in a single call. It is the fastest path when you already have the profile link and just need a way to reach the person.

What does Crunchbase add that LinkedIn and enrichment don't cover?

Funding. A Crunchbase company lookup returns total funding in US dollars, funding rounds with announcement date and type, founding date, employee range and industries from the company's Crunchbase profile. It is the closest public proxy for whether an account can afford what you sell, and a detail neither a LinkedIn profile nor a contact enrichment call provides.

Can I use these contact and company sources over MCP, or only through the API?

Both work. The email finder, email enrichment, email validation, phone enrichment, the LinkedIn lookups and Crunchbase company are all tools on NeuralVerge's hosted MCP server at https://mcp.neuralverge.ai/mcp. Add the URL to Claude, ChatGPT, Cursor or another MCP client, sign in and click Allow. Each tool takes the same inputs and returns the same response shape as its REST endpoint, at the same price.

What does the full five-step sequence cost?

At most 46 points, about $0.046, when every step resolves: 10 for the profile match, 10 for the profile and email, 10 for the finder, 1 for validation and 15 for Crunchbase. Each lookup except validation is charged only when it resolves, so a step that finds nothing adds nothing. Point costs come from the source pages and can change, so check them before you budget a large run.

About NeuralVerge

Give your agents structured, cited, real-world data from 300+ sources and the open web — through one API or MCP server.

Data Sources on the NeuralVerge blog.