Free vs paid code analysis servers routine stays free; specialist work meters.

An agent fixing code needs constant, cheap analysis — find references, check types, run the linter, map the blast radius of a change. Servers like ckb, JavaLens, and lint serve precisely this loop, running against source that's already on disk. Per-call pricing here would be absurd: the calls are constant, the marginal cost is nothing, and the data never leaves your machine.

If a tool analyzes code you have, with compute you have, free self-hosting wins. Full stop. This is most of the category, and it's why most of the category is open source.

Some analyses are rare, deep, and dependent on tooling most teams don't keep installed. Diagnosing a production crash dump through WinDbg, decompiling a Mach-O binary with Ghidra, or running ISO-standard vibration diagnostics like the Predictive Maintenance server are not loops — they're events. The value of one good answer is high, and the setup cost behind it (licensed tools, tuned environments, encoded expertise) is exactly what you're avoiding by not doing it yourself.

These are the listings where an agent paying per analysis is rational: the alternative isn't "free," it's hours of environment setup for a task that happens four times a year.

With x402 attached to a listing, the transaction is one round trip: the agent submits the analysis request, gets 402 Payment Required with the price, settles in USDC on Base in about two seconds, and the analysis runs — payment confirmed before execution, with an Ed25519-signed receipt as proof of exactly what was bought. No vendor onboarding, no seat license, minimum $0.01 per call.

One caution specific to this category: code is sensitive input. Whether free or paid, check what a remote analysis endpoint retains. A signed receipt proves the payment; the operator's documentation has to speak for the data handling.

Keep the inner loop free and local — lint, navigation, type intelligence, impact analysis. Buy the events: crash diagnosis, reverse engineering, domain-specific statistical work where the tooling or expertise isn't yours. If an analysis requires software you'd otherwise have to license and learn, per-call is the cheap option even at dollars per call, let alone cents.

Maintainers whose servers encode rare expertise — WinDbg fluency, Ghidra automation — have the strongest claim to pricing in this category. Claim the listing on Loomal and price the event, not the loop; the 5% fee on settled transactions is currently waived.