Company research with Thirdwatch MCP: discovery to evidence

Company research has several different jobs: discovering businesses in a market, checking a known company and understanding public signals around a competitor. Choosing the right tool prevents a broad search from being mistaken for verification.

Discover businesses by category and place

Use search_businesses when the question is “which businesses match this category in this area?” A small US example is:

{
  "query": "coffee shops",
  "location": "Austin",
  "country": "US",
  "max_results": 10
}

First ask your assistant to call estimate_cost for the tool and geography. Then request a table with name, category, website, address, rating when returned and source links. Leave unavailable fields blank. Google Maps supplies global coverage; Indian discovery may also include JustDial and IndiaMart. Geography affects the source mix.

This is an illustrative query, not a validated list of businesses. A directory entry may be incomplete, duplicated or outdated. A published phone number is not proof of consent to outreach.

Enrich a known company

For an identified business, enrich_company collects company information and supported review sources. Pass company_name and, when known, an explicit company_domain and country. It never guesses a domain from a company name. Without an explicit domain, Trustpilot is skipped and the source charge is lower. Give your assistant the company identity you actually mean: ambiguous names can mix unrelated companies. Check that locations, domains and source links refer to the same entity before summarizing.

For a company whose domain you already know, an illustrative input is:

{
  "company_name": "Stripe",
  "company_domain": "stripe.com",
  "country": "US"
}

Preview this call with estimate_cost, passing tool_name as enrich_company, query as Stripe, and the same company_domain and country. Omit the domain from both calls when it is unknown; do not manufacture one to add review coverage. Inspect warnings and the actual charged credits after execution.

Use verify_business for a business-verification task rather than treating discovery as verification. Registration and directory evidence have different meanings. A successful data lookup is not an assurance of financial health or commercial suitability.

The company research workflow starts with discovery and a deliberate enrichment step. Each paid follow-up is a separate call; review its preview and the actual response metadata.

Research competitive signals separately

competitive_snapshot can combine reviews, hiring, employee profiles, launches and news into source-specific sections. Example:

{
  "company": "Stripe",
  "sources": ["trustpilot", "google_news"],
  "max_results": 10
}

This deliberately narrows the request. Ask the assistant to identify the entity represented by each source, retain original links and keep reviews separate from news. A missing section must stay missing. Ten returned employee profiles would not equal a ten-person workforce, and job listings would not establish revenue growth.

A brand_sentiment sample can help organize review and discussion signals. It is not a representative survey. Make the sources, sample size, time coverage and gaps visible before interpreting a pattern.

Produce an evidence brief, not a confident guess

A useful brief has three parts: returned facts with links, interpretation explicitly labeled as interpretation, and unresolved questions. Include the query and request date so someone can repeat the work. Do not silently turn a missing website into a guessed domain or a missing registration number into a verified business claim.

These tools return results, metadata, warnings and suggested next steps. Read warnings before compiling a polished report. Source failure and a source returning no matches are different outcomes, even if both leave a blank table cell.

Connect Thirdwatch and check current pricing. Thirdwatch account credits serve the MCP workflow. The source-specific Apify tutorials remain useful when you need batch exports or Actor-specific input controls; those use Apify billing separately.