TollNode MCP¶
The missing number in most AI stacks is the price of a call. A business owner can see what a model costs per million tokens and still have no idea what a sales list, an email verification pass or a sanctions screen costs to produce. Every tool arrives with a claim and no invoice, and the invoice is the thing the operator actually has to sign.
TollNode is a gateway that puts 155 tools behind one endpoint with a price printed on each of them, charged in USDC on Base, per call. What earns it a page here is not the count. It is that the price list is a free, machine-readable tool, the payment challenge carries the same number in base units, and the two agree to the cent on every line measured.
Server type: remote Streamable HTTP, stateless
Endpoint: https://tollnode.com/mcp
Server: tollnode, version 1.0.0, title TollNode
Protocol: 2025-06-18
Payment: x402 version 2 on eip155:8453 (Base), asset USDC
at 0x833589fCD6..., or an account API key
Tools: 156 listed; 155 priced; 1 free (list_models)
Website: https://tollnode.com
Listed on: the mcp.so submission queue as #7395
The evidence tier¶
Measured to the payment gate, and stated as such. initialize answered 200 as tollnode 1.0.0 on protocol 2025-06-18 with a full instructions field and capabilities.tools carrying listChanged: false. tools/list returned 156 schemas. The free list_models tool was called and returned all 155 priced tools with price, parameters and category. Three paid tools were then called without any payment and each returned an x402 version 2 challenge, which is the evidence recorded below.
No paid call was executed. Nothing in this guide was purchased. The price claims are checked against the challenge the server itself issues, and the one claim that cannot be checked without paying is named as such in its own section.
The catalogue tool is the free one¶
The arithmetic of the count closes exactly:
| Measurement | Value |
|---|---|
tools/list schemas |
156 |
list_models count field |
155 |
Tools in list_models missing from tools/list |
0 |
Tools in tools/list missing from list_models |
1 (list_models itself) |
155 priced tools plus the free index tool is 156. A caller counting only what it can pay for gets 155, and the difference is the one tool that costs nothing, which is also the one tool a caller should run first.
What the 155 actually are¶
list_models returns a category on every tool, and the categories sum to the count:
| Category | Tools | Category | Tools |
|---|---|---|---|
| Image | 56 | Lookup | 5 |
| Video | 30 | Shopping | 5 |
| Social | 18 | Travel | 4 |
| Audio | 12 | Document | 2 |
| Business | 10 | Jobs | 2 |
| Search | 8 | Text | 2 |
| Property | 1 |
56 + 30 + 18 + 12 + 10 + 8 + 5 + 5 + 4 + 2 + 2 + 2 + 1 = 155. The largest share is generation, which is not what this catalog is for. The ten Business tools are, and they are priced individually rather than bundled.
The ten business tools and what each costs¶
| Tool | Price as published | What it returns |
|---|---|---|
company-contacts |
$0.020 per website | Emails, phones and addresses a company publishes |
company-registry |
from $0.003 per lookup | SEC financials, filings, insider trades |
email-verify |
$0.003 per email | Status, score, MX, mailbox, catch-all, disposable |
maps-business-search |
$0.042 per search | 10 Google Maps businesses with address and category |
google-reviews |
$0.014 per search | 20 newest Google reviews with owner replies |
linkedin-company |
$0.007 per company | One company page as JSON |
linkedin-company-search |
$0.030 per search | 10 companies by keyword and location |
linkedin |
from $0.007 per call | Company profiles, employees, posts, people |
sanctions-screen |
$0.010 per name | OFAC SDN and non-SDN, UK, EU and UN lists |
trustpilot-reviews |
$0.022 per search | 20 newest Trustpilot reviews |
The rest of the surface is priced with the same granularity, and a few lines are worth knowing before a workflow is designed around them:
- Search:
web-search$0.008,url-to-markdown$0.006,news-search$0.035,search-trends$0.037,webpage-screenshot$0.008,website-reader$0.042 for up to 10 pages,web-search-markdown$0.017,googlefrom $0.014. - Jobs:
indeed-jobs$0.006 for up to 25 listings againstlinkedin-jobs$0.016 for the same cap. The same question costs 2.7 times more on one source than the other. - Lookup:
qr-code$0.002,weatherfrom $0.002,domainfrom $0.003,base-wallet-balance$0.003,vin-decode$0.005. - Travel:
tripadvisor-reviewsandbookingfrom $0.053,hotel-search$0.054,airbnb-search$0.056. Travel is the most expensive vertical after generation. - Document:
data-filefrom $0.003 to check a CSV or Excel file before import,document-to-markdownfrom $0.007.
The challenge, and the arithmetic that matches it¶
Calling a paid tool with no payment returns isError and a JSON body, not a refusal to price it. The body names the resource, the method, a description of the call and an accepts array:
sanctions-screen network eip155:8453 amount 10000 -> 10000 / 1e6 = $0.010
email-verify network eip155:8453 amount 3000 -> 3000 / 1e6 = $0.003
company-registry network eip155:8453 amount 5000 -> 5000 / 1e6 = $0.005
USDC carries six decimals, so base units divided by 1,000,000 is the dollar figure, and all three reproduce the published price: sanctions-screen at $0.010 a name and email-verify at $0.003 an email are exact, and the asset is USDC at 0x833589fCD6... on Base with a single payTo address. The quote and the charge are the same number, which is the property that makes a metered gateway usable by an agent: the price is discoverable before the call, not after it.
The instructions state the flow plainly rather than hiding it behind a client library: call the tool, read the payment requirements in the error result, then call again with the signed payment in params._meta["x402/payment"]. The alternative is an account API key on the same endpoint.
The default action costs more than the headline¶
company-registry publishes from $0.003 per lookup and carries a priceOptions block that explains what "from" means:
| Action | Price |
|---|---|
us_filings |
$0.003 |
us_financials (default) |
$0.005 |
us_insider_trades |
$0.008 |
The default action is us_financials at $0.005, which is 67% more than the headline figure a reader will quote. That is not a bait price: the range is printed, and the default is printed. It is the reason a cost estimate should be built from the priceOptions block rather than from the summary line, and it is exactly why the free list_models tool exists.
The gate fires before the arguments are checked¶
This is the finding that a reader should weigh, because it is the one claim on the surface that cannot be verified without money.
sanctions-screen was called three ways: with a real 12 character name, with no arguments at all, and with a 400 character name. All three returned the identical x402 challenge. A malformed call is priced the same as a valid one.
The server's own sentence is You are charged only when a call delivers a usable result, and the challenge body repeats it per tool as Charged only when the call succeeds. That promise is enforced after payment, on the server, and is therefore not observable from an unpaid session. An agent that signs first and validates later pays for the privilege of learning its argument was wrong. The practical rule for a caller is to take the JSON Schema from tools/list, validate locally, and only then sign, because the payment gate will not do it.
Two smaller notes from the same probe: the instructions describe a free list_models filter by category or search word, so the catalogue is queryable rather than a 155 row dump, and the server advertises websiteUrl in serverInfo, which the docs can be checked against.
What a real workflow costs¶
The useful property of a printed price list is that a workflow can be costed before it runs. Enriching 10 companies from a keyword search to a verified email:
linkedin-company-search $0.030 x 1 (up to 10 companies) = $0.030
company-contacts $0.020 x 10 (one per website) = $0.200
email-verify $0.003 x 10 (one per email) = $0.030
total = $0.260
$0.026 per company, which is the sort of number an operator can compare against a data vendor's per-record fee without asking a salesperson. The same arithmetic on a sanctions pass over 500 counterparty names is $5.00 at $0.010 a name, and over 50 names is $0.50, which is a screening run priced in advance rather than discovered on an invoice.
Where it is thin, measured rather than assumed¶
- One vendor, many products. The tool descriptions name ElevenLabs, Google's Lyria, MiniMax, Stability and others, so a defect in one line is not a defect in the gateway, and a price change on one line is not a gateway price change. Read the per-tool price, never a blended one.
- Generation dominates the catalogue. 98 of 155 tools are image, video, audio or social generation. A reader looking for business data should go straight to the Business, Search, Lookup and Jobs categories, which together are 27 tools.
- No free tier beyond the catalogue. Every data tool is metered. There is no anonymous sample call to test a tool's usefulness with, which is why this guide records the challenge rather than a result.
- USDC on Base only for the payment path. An operator without a wallet needs the account key, and the account path is not specified in the instructions beyond the bearer header.
FAQ¶
Does it need a key?¶
No key is needed to connect, list tools or receive a price quote. Every paid call returns a price challenge instead of a result, so an unpaid session is a price enquiry and not a trial.
Can I see the whole price list without paying?¶
Yes. list_models is free, returns all 155 tools with price, priceOptions, parameters and category, and takes a category or a search word. It is the one call to make first.
Is the price I see the price I pay?¶
On the three tools measured, yes: the challenge amount in USDC base units divided by 1,000,000 reproduced the published price exactly. Where a tool publishes "from" a price, check priceOptions, because the default action is not always the cheapest one.
What happens if my call is malformed?¶
You are asked to pay anyway. The challenge fired on empty arguments and on a 400 character name, so validate against the schema before signing rather than expecting the gate to catch a bad argument.
Where does this fit a business?¶
Counterparty screening, company enrichment with email verification, and priced data lookups inside an agent workflow are the three places this beats a subscription whose per-call cost nobody can state. It is the first entry in this catalog for a per-call priced gateway, which makes it the reference for what a metered MCP surface should publish.
Related reading¶
- External MCP server catalog for the rest of the servers evaluated and catalogued here
- mcp-sanctions and Sanctions Screening for the single purpose sanctions surfaces this gateway prices as one tool among 155
- CorpusIQ connectors for the read-only business systems this catalog is built around
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.