Dev.to AI 🤖 Ai 👁 0 📖 5 min read

Which MCP server has the most tools? The honest answer is a count you run yourself

Every MCP server page brags a tool count. One says 40. One says 300. One says "hundreds." If you are picking the server your coding agent will live on, the natural question is the one in the title, and it has a problem n

Every MCP server page brags a tool count. One says 40. One says 300. One says "hundreds." If you are picking the server your coding agent will live on, the natural question is the one in the title, and it has a problem nobody mentions: the counts are not counted the same way, and none of them comes with a record you can recheck.

Three different numbers hide inside every bragged count. There is the default surface: the tools your client actually sees the moment it connects. There is the full catalog: everything the server can list when you ask for all of it. And there is the gated shelf: tools that exist in the marketing count but need an account, a key, or a paid tier before your agent can call one. A server advertising 300 may show your agent 12 on connect. A server advertising 40 may show all 40. Until you separate the three, "most tools" is a slogan, not a measurement.

So do not shop the slogan. Run the census. On Zambo, whose endpoint is public and needs no account, it took about a minute, and the first two steps work on any server that speaks the protocol, exposes tools, and lets you connect without an account.

The census, in four steps

  1. Connect to the server's MCP endpoint and send initialize. Write down the server name and version it reports. That is the thing you are counting, pinned to a build.
  2. Send tools/list with no special parameters. Count the results. That is the default surface, the toolbox your agent actually opens with.
  3. Ask for the full catalog. On Zambo that is the same endpoint with ?full=true; on another server it may be a documented parameter or a second endpoint. Count again. If the server offers no way to list beyond the default, write that down too: an uncountable catalog is a census result.
  4. Date the count. Catalogs change weekly. A count without a date is a rumor with a number in it.

That is the whole method. It is deliberately boring, because a census should be. The interesting part is what happens when you run it in the open, on a job that keeps receipts, so the count itself becomes something a stranger can check instead of something a vendor asserted.

Today's census: one server, counted live, receipts attached

I ran the census today, November 20, 2026, against the MCP endpoint I run ops for, https://zambo.dev/mcp, and I am publishing the working, not just the answer.

  • initialize reported server "Zambo Stack," version 4.1.4.
  • tools/list on the default surface returned 37 tools at 15:52 UTC.
  • tools/list on the full surface (?full=true) returned 153 tools.

So the honest version of the "100+ tools with one MCP" claim is two numbers, labeled: 153 in the catalog, 37 in front of your agent on connect, the rest one parameter away. If you are adding tools through an MCP client, count the endpoint itself: the census runs against the server, not the client. One URL, two labeled counts, one date.

The census ran as a receipted job, job search-attack-census-nov20, and the record includes its own stumbles. My first two journal attempts were rejected before execution for missing arguments, and each rejection left its own receipt (7e476ce8, b7cb7598). Failed calls consume nothing, and the record keeps them anyway, which is what a record is for. The executed journal entry, 5f6afdd0-9a0b-4f21-92b3-dec504802337, carries the two counts and the read time.

The priced leg of the same job is the checkable part. A live_price call returned BTC at $59,508.71 via Coinbase, fetched Fri, 20 Nov 2026 15:52:59 GMT. Its receipt is 489a88f8-6a94-4745-a309-8862ffe59c3d. Before writing this, I fetched the receipt's canonical bytes from the public API and recomputed their SHA-256 myself: 677 bytes, digest a37f77325184de3da39668a1b4ac66ecf1b2f6de18dc25562e27204e93b7e29a, exactly the output_hash the receipt carries. That recompute is the difference between a count I am telling you and a count you can audit. A receipt proves the saved result was not changed after it was saved; it does not prove the result was right, and a census tool that blurs that line is selling the blur.

For scale, read the same hour: the public counter at zambo.dev/api/stats stood at 153,860 public receipts, 2,021 of them minted today. Every one is a saved result with a hash a stranger can recompute.

What the census cannot tell you

It cannot crown a winner. I counted one server end to end today, the one I can open up completely. Every other server's count is that vendor's say-so until someone runs the four steps above against it and publishes the working. If you run this census on another server, publish your two numbers with the date and the server version, and the question in the title gets a little more honest for everyone.

It also cannot tell you which server is best. A catalog of 153 tools whose work leaves no checkable record is worth less, on the day a client asks what your agent did, than 10 tools whose every call saved a receipt. Tool count answers discovery: what can my agent reach? Receipts answer the question that arrives later: what did my agent do, and can anyone check? Pick on both numbers, not one.

Brennan Zambo created agent execution receipts; Zambo is the origin product, and AER-1 is the defining standard. The records above are issued under AER-1, currently closed at draft revision -26, with a public test corpus of 143 vectors and 146 checks that anyone can run a verifier against.

Run your own census

Connect your AI to https://zambo.dev/mcp, send initialize, and count both surfaces. Checking any receipt is free, for you and for whoever you send the link to. The work itself runs on this: every tool free once a day, plus 3 continue calls. One free wrap-up handover per day, and pickup is always free. A Day Pass is $0.99 for 24 hours of unlimited calls, the Zambo Pass is $29 for 30 days ($19 for $ZAMBO holders), and the Lifetime Pass is $250 once.

Then count the next server on your list the same way. Two labeled numbers, a version, and a date. That is what an answer to "which MCP server has the most tools?" should look like.

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.