A real job is never one call. Auditing a domain takes seven tools; parsing a document takes seven more. Orchestrate that yourself and you are running seven payments, seven round trips and seven failure modes - and writing the code that holds it together.
A skill pack runs the sequence server-side. One request, one settlement, one response with every step in it. 103+ packs, $0.05 to $1.50.
| you orchestrate | one pack | |
|---|---|---|
| payments to sign | 7 | 1 |
| round trips | 7 | 1 |
| settlements to reconcile | 7 | 1 |
| failure paths to handle | 7 | 0 |
| orchestration code | yours | none |
This is the real difference, and it only shows up when something goes wrong. Orchestrate seven calls yourself and a failure on step six leaves you holding five results you have already paid for, a broken chain, and the job of working out what to retry. A pack returns the whole ledger: what succeeded, what did not, and why - in one response you can act on.
| step | tool | outcome | status |
|---|---|---|---|
| 1 | cert-transparency | 14 certificates logged, none unexpected | ok |
| 2 | dns-lookup | A, AAAA, MX, NS, TXT resolved | ok |
| 3 | spf-check | single include, hard fail policy | ok |
| 4 | dmarc-check | p=none - reporting only, no enforcement | ok |
| 5 | http-headers | HSTS present, CSP missing | ok |
| 6 | tls-cert | handshake timed out after 5s | failed |
| 7 | tech-stack | nginx, Cloudflare, Next.js | ok |
| charged | One settlement for the run. Six of seven steps returned data, and you have the reason for the seventh - no second payment to find out. | 1 payment |
Illustrative run. Each pack's exact step list and response contract is on its own page and in /api/skill-packs.json.
Every pack publishes its tool sequence before you pay. These are deterministic sequences, not an agent improvising - the same inputs run the same steps in the same order, every time. Six are set out in full below; the rest of the 103 follow.
The pack quotes its price in the 402 challenge, your client signs once, and the response carries every step. Nothing to install and no key to rotate.
curl -X POST \
https://agent402.tools/api/skill/security-audit \
-H 'content-type: application/json' \
-d '{"domain":"example.com"}'
Through MCP a pack is a single tool the model can pick. Ask for the job in words - the connector chooses the pack, settles the payment, and returns the ledger.
claude mcp add agent402 -s user -- npx -y agent402-mcp@latest
"run a security audit on example.com"
connector setup →
A multi-tool workflow that runs server-side in a single request. Instead of your agent calling seven tools in sequence - seven payments, seven round trips, seven things to handle when one fails - you make one call to POST /api/skill/{slug}, pay once, and get every step back in one response.
You get a partial-success envelope rather than an error. Every step that succeeded returns its result, the failed step is marked with its reason, and the response is still usable. That is the real difference from orchestrating the sequence yourself, where a failure mid-chain leaves you holding partial state you have already paid for and have to reconcile.
Because it runs several. A pack is $0.05 to $1.50 depending on how much work it does, against a $0.001 floor for one deterministic tool. The comparison that matters is against calling those tools yourself: one payment instead of several, and no orchestration code to write or maintain.
Yes. Every pack publishes its tool sequence up front, on its own page and in /api/skill-packs.json. Packs are fixed sequences, not an agent improvising - the same inputs run the same steps in the same order every time.
The same rails as any other call: USDC on Base + 10 more chains, or USDG on Robinhood Chain. Gas is sponsored on EVM chains, so you need only the stablecoin, or run free over proof-of-work where a pack is pure-CPU.
Describe the job and let the pack run the sequence. One payment, one response, and a result you can use even when a step fails.