Most people who need a filtered list of companies do not have a database to query against. They have an Apollo or Crunchbase account and a search box that they use manually. The data is there, but there is no easy way to get it out programmatically.
On 19th November 2026, I made three requests to NinjaPear's Company Search API. And I got back three filtered lists of companies. Each request took less than a second and cost me 2 credits, about 2 cents.
Here is the middle call, exactly as it went in and exactly as it came out. The input was four parameters:
| Parameter | Value |
|---|---|
query |
payments infrastructure |
country |
US |
employee_count |
50-500 |
page_size |
10 |
The output was ten companies:
| # | Company | company_details_url |
|---|---|---|
| 1 | SISA Information Security | https://nubela.co/api/v1/company/details?website=http://www.sisainfosec.com |
| 2 | Daon | https://nubela.co/api/v1/company/details?website=http://www.daon.com |
| 3 | Alphapoint | https://nubela.co/api/v1/company/details?website=https://alphapoint.com |
| 4 | Pragmatix Services | https://nubela.co/api/v1/company/details?website=http://www.pragmatixservices.com |
| 5 | Ensim Corporation | https://nubela.co/api/v1/company/details?website=http://www.ensim.com |
| 6 | Paxos | https://nubela.co/api/v1/company/details?website=https://paxos.com |
| 7 | Landmark Global | https://nubela.co/api/v1/company/details?website=http://www.landmarkglobal.com |
| 8 | Aeropost.com | https://nubela.co/api/v1/company/details?website=http://www.aeropost.com |
| 9 | CLS Group | https://nubela.co/api/v1/company/details?website=https://cls-group.com |
| 10 | Performant Corp | https://nubela.co/api/v1/company/details?website=http://www.performantcorp.com |
Every row matches all three filters at once. The response also carries a next_page cursor: call it and you get the next ten matches, and when you are on the last page, next_page is null. This request returned HTTP 200 and cost 2 credits, reported in the X-NinjaPear-Credit-Cost response header. The URLs above are shown percent-decoded for readability; the raw response encodes them.
The usual alternative is exactly that search box. You type your filters into Apollo or Crunchbase, page through the results, and if your plan allows it, export a CSV by hand. And you do it again next week. If you want the same list inside your own pipeline, then programmatic access is another conversation altogether. At Crunchbase, their API is sold via enterprise data licensing. Elsewhere, the search results just don't flow into your code.
By the end of this piece, you can make these very calls to get exact filtered lists in real time.
What a company search API actually is
In plain English, a company search API takes specific filters and returns a list of companies that match your criteria. You can filter by industry, country, city, year founded, employee count, and funding. For every match, the API returns a ready-to-call details URL to fetch more information about that specific company in another step.
It is important to note what this is not. This is not a search engine for company websites. It is also not a LinkedIn data source. NinjaPear does not scrape LinkedIn. The data is compiled from public web sources.
To organize industries, the API uses GICS codes. These are industry classification standards used by financial analysts. For example, the code 45103010 represents application software. If you use a shorter code like 45, NinjaPear will match every industry that falls under the broader category of Information Technology.
How do the filters combine, and where are the gaps?
The filters for NinjaPear's Company Search API are simple. Filters across different parameters use AND logic. This means if you filter by industry and country, it has to be both to show up. Within the same parameter, a comma separated list uses OR logic. If you list US,CA as the country, then the API will return companies from either the United States or Canada. All ranges are inclusive.
There is one important detail here. A field that the dataset does not know about will never match. So if a company has no funding recorded, it will not show up in any funding filter. You should filter on what you need and not add every possible parameter.
I made three calls on 19th November 2026. Each call returned 10 results and cost me 2 credits, about 2 cents.
Call 1 filtered for US companies in application software (GICS 45103010) founded in 2020 or later with over $100M in total funding. Results: Anthropic, Array, Bland, Candescent, Captions, Drata, ElevenLabs, Eve Legal, Federato, and Future Secure AI.
Call 2 filtered for US companies in "payments infrastructure" with 50 to 500 employees. Results: SISA, Daon, Alphapoint, Pragmatix, Ensim, Paxos, Landmark Global, Aeropost.com, CLS Group, and Performant.
Call 3 filtered for companies in the US and Canada that had a Seed or Series A round after 1 January 2026. Results: 4MindsAI, Baselayer, Charm Security, CopilotKit, Dosentra, GKS, HYPRLABS, Ori AI, Orin Labs, and Prefix.
Not all fields are equally covered. Industry, location, company type and founding year are the widest fields. Employee count and funding are thinner. This means that a query that combines four different filters on these two specific fields will return a shorter list than what the raw counts suggest. If you over-filter, your request might come back empty.
What does a search cost, and how do you keep it cheap?
A search request costs 2 credits for every 100 results returned. Given that 1 credit is about $0.01, it means a single request costs me about 2 cents regardless of whether I get 100 companies or just 10. If the result is empty, then the request is free.
There is a free tier available. Accounts in their first 30 days and top 100 subscribers can make one free request every 10 seconds. You can check for the exact charge of any call by looking at the X-NinjaPear-Credit-Cost response header.
To keep costs low, build your query wide first. Industry and location are the widest filters so start there and then narrow down results. You can control the number of results per page with the page_size parameter which ranges from 1 to 100. The default is 100.
If a search returns more than 100 results, you can paginate through them. The API provides a cursor in the next_page field. Every new page that you fetch is a new API call and costs another 2 credits.
How do you call it?
You can get your filtered list with one GET request. Just provide your search parameters as part of the URL, and include your bearer token in the header.
Here is a curl example using the filters from my second call, the one that looked for payments infrastructure companies:
curl -G "https://nubela.co/api/v1/company/search" \
--data-urlencode "query=payments infrastructure" \
--data-urlencode "country=US" \
--data-urlencode "employee_count=50-500" \
--data-urlencode "page_size=10" \
-H "Authorization: Bearer YOUR_API_KEY"
That is the exact request from the top of this piece. The response returns a data object with an array of companies, one name and one company_details_url per match, and the next_page cursor for pagination.
The company_details_url is a ready-made link to everything NinjaPear knows about that one specific company. This is useful because it means the second call to pull out the full record requires no more lookup logic.
Can an AI agent run the search for you?
Yes, and the setup is one URL. In Claude, open Settings, go to Connectors, add a custom connector, and paste in nubela.co/mcp. ChatGPT has a similar flow in its connectors settings: create a custom connector and paste the same URL. That URL is NinjaPear's MCP server. MCP stands for Model Context Protocol, a standard way for an AI agent to call tools. Once the connector is in, the agent can use NinjaPear's 29 tools, and company search is one of them.
Then you just ask: find me US companies in payments infrastructure with 50 to 500 employees. The agent calls the ninjapear_search_companies tool with country=US and employee_count=50-500, and you get the same ten companies my second call returned. No code, no curl, no API key on your side.
The difference from asking the agent to run a web search is the difference between a database and a reading list. A web search returns pages and links. The agent reads them and pieces together what it finds, and the numbers may be stale or wrong. The MCP tool sends your filters to NinjaPear and gets structured rows back: company name, company_details_url, and a next_page cursor for the next ten. The same question gives the same answer every run, at 2 credits a request. One detail if you configure the tool by hand: numeric filters such as funding and founding year must be integers, not strings.
What does the first page of Google sell instead?
If you search for company search tools today, you will mostly find two types of products. First, there are scraping tools that walk through Crunchbase or LinkedIn pages. These are brittle and break when the website changes its front end. Second, there are bulk database vendors who sell a 10GB dump of data for you to filter yourself.
The difference is that NinjaPear sells company search as a service. You pay 2 credits per request to get exactly what you need. You get a details URL for every company and no database to maintain. If you are also weighing enrichment vendors for the rest of your stack, I keep an updated comparison of company data enrichment tools, free and paid.
Three requests, three filtered lists, costing around 2 cents each. Filter, call, and move on to the work the list was for.