nvNeuralVerge

Building an M&A Due-Diligence Agent With a Research API

An M&A due diligence API workflow for an agent: cited first-pass research on a target over REST, with MCP tools for registries and sites. Humans decide.

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

▸On this page

An M&A due diligence API workflow hands the repetitive first pass to an agent. It takes a target name, asks a research platform a fixed set of questions, and returns a briefing in which every fact carries a citation, for a human deal team to check and build on. Research runs over REST; extraction and registry lookups can also be MCP tools the agent uses around it. This is research support only: it gives no legal or financial advice and reaches no conclusion about price or risk.

The short version

  • —What it does. Resolves the target's legal identity, collects ownership, funding and reputation facts from public sources, cross-checks them and cites each claim. Disagreements go into the memo, not out of it.
  • —Who it is for. Corporate development teams screening target lists, investors triaging inbound deals, and analysts building the first-days briefing.
  • —What it does not do. It cannot see a data room, audited financials or contracts, and it does not give legal or financial advice. People and their advisors make every judgment call.
  • —What it costs. AI research is priced by depth tier, from 20 points for Lite to 400 for Ultima. Search and AI extraction are 5 points per call. Plans start at $50 a month for 50,000 points; see pricing.
  • —How you reach it. Research is REST only. Search, AI extraction and every data source are also MCP tools.

What an M&A due diligence agent should do

Some diligence work is mechanical: confirm the entity exists, list its officers, pull its funding history, summarize public reviews. The rest is judgment: whether a customer concentration is acceptable, whether a clause is a problem, whether the price is right. An agent earns its keep on the first kind. The second stays with people.

Work worth handing to an agent

Drawing the line by source works well. From public information an agent can confirm the target's legal identity, status and officers from official registers; assemble funding rounds, investors and stated headcount; summarize public review sentiment; read the target's own website into structured fields; and cross-check all of it, flagging where sources disagree.

Work that stays with people

A public-source agent cannot see a data room, audited financials, contracts, or anything a target has not published, and it cannot value the company or advise on legal exposure. State that boundary in every memo.

Why one search call is the wrong tool

The instinctive first version wires a web search into an LLM and asks it to "research the company." The result is fluent paragraphs built on links, not facts, and someone still has to open each page and reconcile it with the others. In diligence the gaps matter more than the fluency: a stale ownership record, a subsidiary under a different name, a funding round reported differently in two places.

What the agent needs is a capability that plans a compound question, routes each part to the right kind of source, extracts the facts, checks them against each other and cites every claim. That is what NeuralVerge's AI research does. The agent supplies the workflow and the questions; the platform does retrieval and verification.

The workflow in four stages

Keep four stages separate so each can be tested alone: intake, screen, investigate, brief. Splitting screen from investigate is the main lever on turnaround and cost, because most candidates should drop out before the expensive pass.

Stage 1: intake and identity

Most diligence errors start with the wrong entity: names collide across countries and trading names differ from legal names. The intake record should carry the trading name, likely legal name, country of registration and website.

Take Acme Oy (Finland), a fictional example: trading name Acme, legal name Acme Oy, country Finland, website supplied by the deal team. The first question is narrow: "Is Acme Oy (Finland) an active registered company, and what is its registered legal form?" That routes to a national corporate registry and is quick to answer. If the result is ambiguous, say two entities with similar names, the agent stops and asks a human instead of guessing. Everything downstream depends on this step.

Stage 2: the screening pass

The screening pass exists to drop candidates quickly. Its questions usually have one authoritative source:

The usual questions: is the entity registered and in good standing, what is its legal form and registration date, who are the registered officers, and does the company's own website describe the business the thesis assumes?

Because these are narrow questions with clear sources, a lighter research tier is the sensible default. AI research has five depth tiers (Lite, Base, Core, Pro and Ultima), and a Lite run typically returns in 30 to 90 seconds. Running exhaustive research on every name in a long list would be waste.

Where the target has a website, the agent can also use AI extraction to read the company's pages into structured fields: products, locations, leadership names, stated customers and contact details. That gives the memo the target's self-description to compare with independent sources. A call for the Acme Oy example looks like this:

curl -X POST https://api.neuralverge.ai/functions/v1/run-extract \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://acme.example",
    "instructions": "Extract the company name, products, office locations, leadership names and contact details.",
    "settings": {
      "country_code": "fi",
      "extract_schema_json": "{ \"type\": \"object\" }"
    }
  }'

The extracted fields come back directly in machine, ready to diff against the registry record from the identity check:

{
  "session_id": "7a9c2e10-...",
  "kind": "extract",
  "url": "https://acme.example",
  "machine": {
    "company_name": "Acme Oy",
    "products": ["Example product line"],
    "locations": ["Helsinki, Finland"],
    "leadership": ["Jane Example, CEO"],
    "contact_email": "hello@acme.example"
  }
}

The output is a pass, fail or needs-review status per candidate, with citations. Failures drop out, ambiguous cases go to a human, and survivors move on.

Stage 3: the investigation pass

For a shortlisted target the agent asks compound questions that need different kinds of sources and may produce conflicting answers:

  • —Ownership and control: registered owners and officers, and recent changes.
  • —Funding history: rounds, investors and dates.
  • —Market position: how public analyst and category pages describe the company.
  • —Reputation: what customers say publicly, and recurring complaints.
  • —Consistency: whether the company's claims match the registry and third-party sources.

The agent does not have to decide which website answers which question. The platform plans the request into sub-questions and routes each to the right part of the source catalog: registries for ownership, company-intelligence data for funding, review platforms for reputation, the open web for the rest.

Asking questions the pipeline can answer

Question quality decides output quality. "Do due diligence on Acme" yields a vague answer; a fixed question set, one entry per topic, each naming the entity precisely and asking for facts, does not. Ask for "the registered ownership and officers of Acme Oy (Finland), citing the register" rather than whether the company is a good target, and store the questions as templates so every candidate gets the same set.

Cross-checking is the point

Diligence is largely about finding where things do not line up. If a registry lists one set of officers and the website lists another, or a funding database reports a round the company never mentions, that difference should reach the memo as an open question with both citations. A tidy memo that hides conflicts is worse than a messy one that shows them.

Stage 4: the brief

The agent assembles a memo with a section per topic, each factual sentence keeping its citation, and a closing list of open questions for the human team.

A worked example: one target from intake to memo

Take Acme Oy again, with a thesis that assumes a small, profitable software vendor with a couple of institutional investors. The identity question returns a registered, active private limited company with a named managing director and board, cited to the register. Extraction on the site returns a product description, two office locations and a leadership page. The site's leadership list and the register's officer list overlap but differ; the agent records a note rather than a failure, because a site listing a non-statutory executive is normal.

On the deeper pass, ownership routes to registry sources, funding to company-intelligence data, reputation to review platforms, and the consistency question compares the site's stated headcount and founding year to third-party sources. The memo ends with open questions: the officer mismatch, a funding-date discrepancy, and the fact that financial statements were not part of this pass and need to be requested. An analyst clicks through the citations that matter and decides whether Acme moves forward. The agent saved the assembly time. It did not make the call.

REST and MCP

NeuralVerge hosts an MCP server at https://mcp.neuralverge.ai/mcp. Add that one URL to an MCP client such as Claude, ChatGPT, Cursor, VS Code or Gemini CLI, sign in through the browser and click Allow. The Claude walkthrough shows the steps, and guides for other clients are at docs.neuralverge.ai.

The part to be precise about is what MCP covers. AI research is not an MCP tool. It runs over REST, because a research run is asynchronous. What an agent gets as MCP tools are Search, AI extraction (run_extract) and every data source in the catalog, which is where the mechanical lookups live. Tool names use underscores where REST paths use hyphens. Every tool returns the same response shape as its REST endpoint and costs the same, so MCP is an integration choice, not a pricing one.

Research over REST

Research runs as a session: you start it, then poll until it finishes. A screening-tier call for the Acme Oy identity question looks like this:

curl -X POST https://api.neuralverge.ai/functions/v1/run-research \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "instructions": "Is Acme Oy (Finland) an active registered company, and what is its registered legal form? Cite the register.",
    "settings": {
      "country_code": "fi",
      "search_enabled": true,
      "deepsearch_model": "base"
    }
  }'

That returns a session_id right away, which you poll:

curl "https://api.neuralverge.ai/functions/v1/get-session-status?session_id=7a9c2e10-..." \
  -H "Authorization: Bearer YOUR_API_KEY"
{
  "session_id": "7a9c2e10-...",
  "status": "complete",
  "results": {
    "kind": "deepsearch",
    "human": "# Research summary\n\n- Acme Oy is an active Finnish private limited company (osakeyhtiö), registered in 2019. Source: Finnish Trade Register.",
    "machine": {
      "entity_status": "active",
      "legal_form": "Private limited company (Oy)",
      "registered_year": 2019
    }
  }
}

For the deeper pass, the same endpoint takes a heavier deepsearch_model ("core" rather than "base") and, where it matters, a finalizer_model to choose which model writes the final memo text.

Registry lookups and extraction over MCP

The identity check on Acme Oy can also be a tool call. The Finnish company search is run_fi_companies_search:

{ "tool": "run_fi_companies_search", "arguments": { "keyword": "Acme Oy", "max_items": 2 } }

It returns matching companies with Business ID, legal form and home municipality; the Finland company search page is the reference for fields and billing. Reading the target's site is the same run_extract call shown in stage two, with url, instructions and settings as arguments.

Putting them together

A backend job that runs the four stages in a fixed order calls everything over REST, including research, so each stage has a known input and output and you control retries in code. An analyst in Claude or Cursor uses the MCP tools for registry lookups and site extraction mid-conversation, and asks for research through a small tool of your own that wraps run-research. Bound that agent: a maximum number of calls per target, a maximum depth tier, and a rule that every claim in the memo carries a citation returned by a tool. Field-level detail is in the API documentation.

Cost and depth by stage

Depth is a per-request setting, so the agent can match it to the stakes. Points per call, from the pricing page at the time of writing:

  • —Screening a long list: Lite (20 points) or Base (50) research, plus a registry search at 1 point per returned record for Finland.
  • —Reading the target's site: AI extraction, 5 points.
  • —Investigating a shortlist: Core research per question, 100 points.
  • —Deep dive on a final candidate: Pro (200) or Ultima (400) for ambiguous questions.

As an illustration, screening 100 candidates with one Lite research question each is 100 times 20, or 2,000 points, before extraction. A five-question Core set on three survivors is 1,500 points. This is arithmetic on published rates, not a forecast for your question set. Run time grows with the tier: Lite is typically 30 to 90 seconds, Ultima 5 to 12 minutes.

Two limits to plan for. A call whose input does not resolve can still be billed; AI extraction on a dead URL returned null and cost 5 points, so validate URLs before extracting. And a registry record is only as current as the register's last update, so put the run date in the memo and treat it as a snapshot.

How to decide

Before building this, check the design against your own deal flow:

  • —Is your deal flow public-source heavy? Screening long target lists and triaging inbound deals benefit most.
  • —Do you cover the jurisdictions you need? Check the source catalog for your usual target countries and note which have registry coverage.
  • —Can a sample answer show its citations? Confirm that every claim in a test answer links to a specific source, not just a list at the end.
  • —Does a messy target surface its conflicts? Run one you know has awkward public data and check that disagreements appear in the answer.
  • —Is a human reviewing the memo? If nobody will open the citations, the agent is producing prose, not diligence. Keep citations attached to claims, show disagreements instead of resolving them, state the scope, and keep words like "low risk" or "fairly priced" out of an automated memo.
  • —Fixed pipeline or agent with tools? A backend job suits repeatable screening; tools suit an analyst exploring one target.

Frequently asked questions

Can an agent replace a deal team's due diligence?

No. The agent handles the mechanical first pass: finding, extracting and cross-checking public facts about a target. Judgment calls, legal review, financial analysis and anything that depends on non-public information such as a data room stay with the humans and advisors running the deal.

Does this cover financial statements or a virtual data room?

No. The pipeline reads public sources such as registries, funding data, review platforms and the open web. It does not see a target's private financials or a virtual data room. Those documents are reviewed separately, and the agent's job is to make sure the public picture is assembled and checked before that work starts.

How do you keep the agent from stating things it cannot support?

Every claim in a research answer carries a citation to the source it came from, and the pipeline's cross-check step surfaces disagreements between sources instead of hiding them. Your agent should carry both forward: keep the citations in the memo, and flag open conflicts for a human reviewer rather than resolving them silently.

Can the agent monitor a target between signing and closing?

Yes. It is an ordinary call, so you can re-run the same questions on a schedule and compare answers over time. A change of control, a new registry filing or a shift in public reputation is worth a human look, and a scheduled re-check is a way to notice it without anyone remembering to look.

Can I run M&A due diligence research over MCP, or only through the API?

Partly. AI research, the deep pipeline that produces the cited answers, runs over the REST API only. NeuralVerge's hosted MCP server at https://mcp.neuralverge.ai/mcp exposes Search, AI extraction and every data source as tools, so an agent can look up registries and read a target's website over MCP and call research over REST. Same response shapes and same prices either way.

Which jurisdictions does it work for?

It works where the source catalog has coverage, which today includes several national corporate registries alongside funding, reputation and open-web sources. For a target in a country without a supported registry, the pipeline falls back to the open web, and the citations show which kind of source backs each claim. Check the source catalog for the countries your deals actually touch.

How many points does a first pass on one target cost?

It depends on the depth tier you pick. A Lite research run is 20 points, Base is 50, Core is 100, Pro is 200 and Ultima is 400. Search and AI extraction are 5 points per call, and a Finland registry search is 1 point per returned record. Add up the calls your question set makes for one target to get a per-target figure.

About NeuralVerge

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

AI Deep Research on the NeuralVerge blog.