Pricing

The plans above watch a list you choose. If you need the whole dataset instead, there is a Feed API: a cursored, catalog-wide stream of every event we record, and the full snapshot history behind any server.

New servers appearing in the registry, tool definitions changing, endpoints going down and coming back. Resume from a cursor and you get every later event exactly once, and the feed tells you when a gap sits behind your cursor rather than resuming as though nothing was missed.

This is for the gateways, registries, and platforms rebuilding this dataset internally. It is priced by contract, not from this page, because the shape depends on what you are doing with it.

Everything in the catalog is also readable from an MCP server and a JSON API, both open without an account. An agent picking which MCP server to install can ask whether it answers, what its tools claim to do, and whether that claim has changed lately, before it commits to the dependency.

Free either way. A key raises the daily limit, and lets an agent start monitoring a server on your account the moment it picks one.

Most of the servers you depend on will not change this month. In a sample week, the servers that drifted were overwhelmingly ones nobody uses: popular servers are roughly 99% stable week to week.

That is the honest case for this product, not an argument against it. Toolpin is insurance. You are paying to find out on the day it happens rather than the day something breaks in production, and the rest of the time to have it confirmed that nothing changed.

That confirmation is a real deliverable: once a month we send an assurance receipt listing every server we checked, how many times, and the pinned hash it still matches. Quiet months produce evidence, not silence.

What counts as a change?

Any difference in what a tool claims to do: its name, description, input or output schema, or annotations. We hash the fields the MCP specification defines and ignore vendor extras, so a build stamp that changes on every deploy does not wake you up. The exact rule is on the methodology page.

Do you ever call my tools?

No. We perform a standard handshake and read the tool list. We never invoke a tool, never send credentials, and never execute anything a server returns.

Do you tell me whether a server is safe?

Never. We publish observed facts: changed, unreachable, last commit date. No scores, no ratings, no safety judgments. The judgment stays with you.

I need more than 25 servers.

Email corey@macroalchemy.com. If you want the whole dataset rather than a watchlist, that is the Feed API above.

What if my server needs authentication?

We record it as reachable but auth gated, and track its reachability. We do not hold credentials, so we cannot hash the tool list behind an auth wall.