Data Tokens and Usage
Understand how Crunchbase MCP uses data tokens, estimate consumption, and monitor your allowance.
Crunchbase data tokens measure the data and Expert work used to answer your MCP requests. The tokens you use come out of your allowance.
Your AI provider measures text tokens separately. A long company description can cost the same number of Crunchbase data tokens as a short company name.
From your prompt to usage
Your assistant may use several tools to answer one prompt. For example, it might find companies, retrieve their profiles, and then analyze their funding. Each data request adds to your usage, even if your assistant combines the results into one short answer.
What uses data tokens?
| What your assistant does | How usage is calculated |
|---|---|
| Searches for companies, retrieves profiles, or finds name matches | 10 tokens per non-empty basic field value returned. |
| Asks Expert a question or asks it to build or revise a search | Data used for the task + processing + values in the response. |
| Uses Expert to identify a company, person, or other entity | Data used for the task + processing. If a profile is returned, its fields add to the charge. |
| Requests domain guidance or a Knowledge Center answer | Processing tokens are charged. Values in the guidance response do not add a separate charge. |
| Reads tool reference information, checks the token balance, or finds, reads, creates, or updates saved lists | 0 tokens for these operations. Retrieving profiles for companies in a list uses tokens separately. |
| Sends feedback | Positive or neutral feedback uses 0 tokens. Negative feedback that produces a revised answer or search uses tokens for that revision. |
For tool names and parameters, see Developer Reference.
Estimate a search or profile lookup
For basic fields, use this calculation:
Records returned × non-empty fields per record × 10 = data tokens
Suppose a search returns 20 companies with 6 non-empty fields each:
| Returned data | Calculation | Data tokens |
|---|---|---|
| 20 companies, 6 fields each | 20 × 6 × 10 | 1,200 |
| The same response, with 5 field values missing | 1,200 − (5 × 10) | 1,150 |
| Another page with the same amount of data | 1,150 + 1,150 | 2,300 across both pages |
A field is one item of information, such as a company name, website, or funding amount.
Keep these rules in mind:
- Missing values, blank text, and empty lists use 0 tokens.
- A value of 0 or false still counts as a returned value and uses 10 tokens.
- A non-empty list inside one basic field counts once. For example, a basic field containing several industry labels uses 10 tokens.
- Related records, such as funding rounds returned with a profile, have their own fields. Those fields also count.
- Usage reflects the data returned to your assistant. Your assistant may show you only part of that data.
Understand Expert usage
Expert can retrieve data, analyze it, and return an answer or search. Its usage has three parts:
| Part | What it covers |
|---|---|
| Data retrieved | Crunchbase data used to complete the task. Its token charge is included once in the Expert total. |
| Processing | Work needed to understand the request, analyze the data, and prepare the result. The amount varies by request. |
| Response data | Values in the Expert result, charged at the rates below. |
Rates for values in the response
For Expert answers, the type of calculation sets the rate for each counted value in the response:
| Expert answer | Tokens per response value |
|---|---|
| Results without an aggregate or window calculation | 10 |
| Results with a window calculation, such as each company’s rank | 20 |
| Results with an aggregate calculation, which combines records into a count, sum, or average | 100 |
The rate applies to all counted values in that Expert answer, including names returned with the calculated values. If the query includes both window and aggregate calculations, use the 100-token rate.
For example, an aggregate query that returns an industry name and a company count has two response values: 2 × 100 = 200 tokens for the response values. Data retrieval and processing add to that amount.
When Expert builds a search, the returned search definition and title use 10 tokens per counted value. If results are included as a preview, their basic fields add to the charge. A search revised through feedback does not automatically include a results preview.
Expert example
Suppose an Expert answer uses the 10-token rate and has these charges:
Data retrieved
Processing
30 values × 10
This example shows how the parts add up. The charge for an Expert question varies with the data and work needed.
The 200,000-token maximum
Each Expert response has a maximum charge of 200,000 tokens for data retrieval, processing, and response values combined.
- A calculated charge of 1,200 stays 1,200.
- A calculated charge of 250,000 becomes 200,000.
The maximum reduces this charge by 50,000 tokens. It does not raise smaller charges to 200,000.
A results preview or profile returned with the Expert response can add tokens above that maximum. For example, an Expert response charged at the maximum plus a 1,200-token preview totals 201,200 tokens.
Plan your allowance
Basic searches are easiest to estimate from the number of records and fields returned. For workflows that use Expert, first measure a few examples of the work you expect to repeat.
- Check your remaining balance.
- Run several examples of the workflow you expect to repeat.
- Measure the tokens used and calculate the average.
- Divide your remaining allowance by that average.
For example, three workflows use 1,800, 2,400, and 1,800 tokens. The average is 2,000. With 100,000 tokens remaining, that suggests about 50 similar workflows.
Use this estimate to plan. Different questions, extra pages, and more profile lookups can change usage. A prompt such as “use no more than 2,000 tokens” does not set a limit that Crunchbase enforces.
Keep usage focused
- Start with a small result set. Ask for 10 companies before requesting a larger list.
- Request the fields you need. “Return the name, website, and total funding” is more focused than “give me every detail.”
- Add details as needed. Retrieve full profiles only for the companies you want to investigate.
- Reuse data already in the conversation. Ask your assistant to use those results when you do not need a new lookup.
- Check usage before a large task. Measure a small sample before you repeat the workflow for many records.
A shorter final answer does not necessarily use fewer data tokens. Asking for one sentence can still trigger a large data retrieval or a complex analysis. Repeating a request is not guaranteed to be free.
Check your balance
View your MCP usage in your Crunchbase MCP account settings.
You can also ask your assistant:
Check my remaining Crunchbase data-token allowance and reset time.
Your assistant can check this information without using data tokens. The reset time and allowance depend on your account.
Building an integration? See Data-Token Metering.
Common questions
What happens when my allowance runs out?
When your balance reaches zero, Crunchbase blocks additional requests that require data tokens. Check your account for its reset time.
A request that starts with tokens available can finish and use the remaining balance. For example, if a response calculates to 1,200 tokens and 500 remain, Crunchbase normally deducts the remaining 500 and marks the allowance as used up. This does not guarantee how much work a small balance can support.
Do empty results use tokens?
A successful basic search with no returned field values uses 0 tokens. An Expert request can still use tokens for data retrieval and processing even when it finds no matching results.
Do failed or canceled requests use tokens?
Ordinary tool failures that return no usable data use no MCP data tokens. Partial results, completed work before a cancellation or disconnect, and metering errors can still involve a charge. Check your account usage when the outcome is uncertain.
For more detail, see Reliability, Errors, and Pagination.
Updated 2 days ago