cd /news/ai-agents/we-graded-every-remote-mcp-server-in… · home › topics › ai-agents › article
[ARTICLE · art-174173] src=nestdaddy.com ↗ pub= topic=ai-agents verified=true sentiment=· neutral

We graded every remote MCP server in the official registry. Here is what we found.

A daily audit of the Official MCP Registry found that 55.9% of remote MCP servers answered and listed their tools publicly, 33.9% answered but required a key, 9.3% were down and 0.9% needed setup, based on checks of 41,089 gradeable servers out of 41,645 listed as of 22 November 2026. The report assigned grades from A+ to F, with 17.2% of servers earning A+ and 39.4% landing at B, and measured a median 253 ms time to answer the first initialize call. It also counted 322,242 tools across 22,576 servers with public tools, finding 93.3% of tools carry a description of 80 characters or more while 73.6% of servers have no icon in the registry.

recommendation

List your MCP server's tools without a key and require one only to call them, since 33.9% of servers hide tools behind auth and clients cannot show capabilities before connecting.

more in playbook: Designing agent systems
read5 min views3 publishedNov 22, 2026
We graded every remote MCP server in the official registry. Here is what we found.
Image: Nestdaddy (auto-discovered)

A data report on the remote MCP servers in the Official MCP Registry, checked every day: how many answer, how fast they are, how many hide their tools behind a key, how well tools are documented, and the small fixes that move a server from B to A+.

Every day we connect to every remote MCP server listed in the Official MCP Registry and run the same public check: does it answer, how fast, which tools does it list, and how well are they described. No credentials, no tool calls, just what any client sees when it first connects.

From those checks every server gets an MCP Score, a grade from A+ to F. This is what the data said on 22 November 2026.

Press Run to call the API.

Get your free API key

The numbers at a glance #

  • 41,645 remote servers listed,41,089 with enough daily checks to be graded.
  • On the latest check, 55.9% answered and listed their tools publicly,33.9% answered but required a key,9.3% were down, and0.9% needed setup.
  • Median time to answer the first initialize call:253 ms . The fastest tenth answer in under99 ms ; the slowest tenth take more than669 ms .

Grades #

Grade Servers Share
A+ 7,051 17.2%
A 8,457 20.6%
B 16,185 39.4%
C 5,113 12.4%
D 497 1.2%
F 3,786 9.2%

More than a third of remote servers earn an A or A+. The biggest group sits at B, and most of them are one or two small fixes away from A.

Reliability: most servers are up, a long tail is not #

86.8% of servers answered every single daily check. 11.0% answered fewer than nine checks in ten. For an agent that depends on a tool, that tail matters more than speed: a tool that is missing one day in ten breaks a workflow one day in ten.

A server that is down on its latest check is capped at C until it answers again.

Speed: most are fast enough, a few are very slow #

Median first answer is 253 ms. 23.5% answer in under 150 ms, which earns full speed marks. Only 2.2% take longer than 1.2 seconds. Speed is measured as the median of the last seven answered checks, so one slow morning does not sink a server.

A third keep their tools behind a key #

13,848 servers (33.7%) answer but only list their tools after you sign in. That is a reasonable choice for paid or private services, and our grade does not punish it: those servers are graded on reliability, speed and listing, and can still reach A+. But it means clients and directories cannot show what the server does before you connect.

Tip: if you can, let clients list your tools without a key and only require one to call them. People decide faster when they can see the tools.

Tool documentation is better than you might think #

Across the 22,576 servers with public tools, we counted 322,242 tools, a median of 7 per server and up to 80.

  • 93.3% of tools have a description of 80 characters or more.
  • 73.7% of servers describe every tool.
  • 80.7% of the1,134,578 tool inputs have a description.

Good descriptions are not cosmetic. The model reads them to decide which tool to call and what to pass. An undescribed input is a guess.

Tools change often #

25.1% of servers rewrote at least one tool in the last 30 days. Change is healthy, but an agent that learned your tools last week can break when a name or an input changes silently. Our score rewards stability, and every server page shows a history of tool changes with an RSS feed, so users can see what changed and when.

Listing gaps are the easiest wins #

  • 73.6% of servers have no icon in the registry.
  • 6.2% have no source or website link.
  • 1.3% have a description that is too short to be useful.

Each of these takes minutes to fix in your server.json, and directories that read the registry pick it up automatically.

How to move from B to A+ #

From what we see across the registry, the fixes that move grades the most are:

  1. Describe every tool and every input , in plain words, 80+ characters for each tool.
  2. Add an icon, a website and a source link to your registry entry.
  3. Keep tool names and inputs stable. Add new tools instead of renaming old ones.
  4. Stay up. A cheap health check and an alert beat any optimisation.
  5. Answer initialize fast. Do heavy work lazily, after the handshake.

Check your own server #

Find your server at nestdaddy.com/mcp/servers, or look it up on the MCP Score page. Every server page shows the grade, each part of the score with tips, the full check history, and a button to check it again now after you ship a fix.

If you like your grade, every server has a live badge for its README:

[![MCP Score](https://nestdaddy.com/mcp/servers/<your-server>/score.svg)](https://nestdaddy.com/mcp/servers/<your-server>)

The badge updates by itself after every daily check. It is free, the directory is free, and it always will be.

Method #

  • Source: every remote server in the Official MCP Registry, synced daily.
  • Check: one public connection per server per day:initialize , thentools/list . No credentials, no tool calls.
  • Score: Reliability 30, Speed 20, Tool docs 25, Stability 15, Listing 10. The percentage is over the parts we can measure, so a server that requires a key is graded on reliability, speed and listing. Grades: A+ from 95%, A from 88%, B from 78%, C from 65%, D from 50%. A server needs three daily checks before it gets a grade.
  • Numbers: taken from the sync that finished on 22 November 2026. The live figures on each server page change every day.

Full method: nestdaddy.com/mcp/score.

── more in #ai-agents 4 stories · sorted by recency
── more on @official mcp registry 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/we-graded-every-remo…] indexed:0 read:5min 2026-11-22 · —