Audit ledger · Generated from the live database · October 31, 2026

What reviewing 2,333 Claude tools actually looks like.

The Claude Observatory tracks 4,102 tools — skills, MCP servers, hooks, patterns, and workflows. 2,333 of them have reached a verdict. This page is the ledger: every number below is queried from the review database the moment the site builds, so it can't go stale and it can't be massaged.


1,924 Approved

Published with a grade, a review depth, and caveats where they're due.

377 Rejected

Didn't clear the bar. They stay in the database — the no's are part of the record.

32 Security holds

Frozen pending a security question we couldn't resolve. Not listed until it is.

01The full ledger

Every tool, every status

1,749 tools are still in the pipeline — evaluating or staged for triage. They don't get listed until they get a verdict.

  • Approved · 1,924 · 46.9%
  • Rejected · 377 · 9.2%
  • Security hold · 32 · 0.8%
  • Evaluating · 1,749 · 42.6%
  • Deprecated · 20 · 0.5%

Of the 2,333 tools that reached a verdict, 17.5% didn't make it — roughly one in 6. A catalog that approves everything isn't a review; it's a directory.

02Grades of the approved

Approval is not endorsement

Each approved tool carries a letter grade from its latest evaluation — a weighted score across client readiness, breadth of use, reliability, and security.

Grade A Recommended without hesitation 975.0%
Grade B Solid — minor caveats 51026.5%
Grade C Usable — know the limits 1,13859.1%
Grade D Approved with warnings attached 1538.0%
Grade F Kept only for the record 261.4%

Grade C is the biggest bucket at 59.1% of approvals. Only 97 tools have earned an A. That's the honest shape of this ecosystem right now: mostly usable, rarely exceptional.

03Review depth

How closely each tool was examined

Not every review is the same review, and pretending otherwise would be dishonest. Every listing on the site discloses its depth.

Tested Installed and run hands-on, end to end 130.7%
Reviewed Source and docs read closely, not run 1789.3%
Scanned Automated signals + structured pass 1,72489.6%
Listed Catalogued with basic metadata only 90.5%

Only 13 tools have been personally tested end to end; 89.6% are scanned. Hands-on testing is the scarcest resource in this catalog — which is exactly why we label it.

04Domain skew

Where the catalog leans

Approved tools across the eight domains we track. The skew is real, so we show it.

Code & development 1,03253.6%
Productivity & workflow 22611.7%
Data & analytics 1568.1%
Documents & content 1336.9%
Security & compliance 1135.9%
Config & setup 1015.2%
Infrastructure & DevOps 1005.2%
Communication & collaboration 633.3%

Code & development alone is 53.6% of everything approved. The Claude tool ecosystem still builds mostly for developers — a gap worth knowing about if you're shopping for anything else.

05Trust signals

What the factual record shows

Alongside reviews, the pipeline collects factual signals from GitHub and npm for 1,924 of the 1,924 approved tools. Three that matter:

85.6% Known license

1,646 approved tools have an identifiable license. The rest — you're deploying on trust.

87.7% Commit in last 90 days

1,688 approved tools show recent maintainer activity as of the latest scan.

0 Known CVEs recorded

Across all approved tools' registry records at last scan. Absence of a CVE is not a security guarantee — see review depth above.

06The work behind it

The paper trail

Evaluations logged Append-only scoring history — re-reviews included 7,217
Field notes Tried it; here's what happened 179

Trust-signal scans for approved tools run continuously: the oldest current scan dates to October 21, 2026, the newest to October 31, 2026. Tools that fail on re-review get downgraded or moved to the not-recommended list — the grade you see is the latest, not the best.

How to cite this

This page is built to be referenced. Cite it as:

Matthews, A. "The Dataset Report: What reviewing 2,333 Claude tools actually looks like." Value Alignment Consulting, October 31, 2026. https://valuealignmentconsulting.com/dataset-report

Numbers regenerate from the live review database on every site build, so figures on this page move as the catalog grows. The methodology is documented on the evaluation guide and about pages.

Rolling Claude out in your org? Let's talk.

Start a conversation →