How many DeepSeek Harness plugins? 3 counts explained

DeepSeek Harness does not have one plugin total. Count repository entries and you get one number; count deduplicated slugs and you get another; count searchable Hub listings and you get a third. All three are correct answers to different questions.

Three numbers first: 19,783, 16,079, and a range

As of 2026-11-01, the site plugin index contains 19,783 repository entries and 16,079 deduplicated plugin slugs. Both come from the same snapshot; the only difference is whether you count repositories or names.

The gap between 19,783 and 16,079 comes entirely from same-name duplicates. 1,705 slugs map to two or more repositories, producing 3,704 extra entries. That is not a data error; it is what a community ecosystem looks like.

You cannot derive the Hub number from the first two. The Hub has its own inclusion and indexing rules; a repository may exist without being searchable. Use the market on the day for searchable counts, and the first two numbers for ecosystem size.

Why one name maps to dozens of repositories

Duplicate names are not rare; plugin naming has no global uniqueness constraint. Authors choose names, and generic names like dsh-memory or dsh-session-delete naturally attract multiple implementations.

The most duplicated names in the index:

The more duplicates a name has, the less you should choose by name, and duplicates also signal wildly different maintenance quality. These ten names alone account for more than 300 repositories. Searching by name hands the decision to ranking; owner, recent commits, and compatible version are the real filters. Some are complete implementations; others are experiments or forks. A simple check works: look at the last commit, whether the README documents installation and compatibility, and whether issues get replies.

Category distribution: where supply concentrates

Split the 19,783 entries by category and the weight becomes visible. Categories come from the index directory structure; plugins in one category solve similar problems.

Tools account for nearly 40 percent, which shows that "give the agent more abilities" is the dominant motivation. Interface and model follow: one changes interaction, the other changes model supply. Memory and workflow are small but targeted.

Category counts are not usable counts. Tools has the most duplicate names, so deduplication narrows the gap; memory is small but heavily duplicated, making selection harder. Read entry count and duplicate density together.

Why the three calibers keep diverging

The gap is not a one-time difference; it keeps changing with the ecosystem. Three reasons, all structural.

1. Repositories keep growing. New authors publish new plugins, raising entry counts; duplicate names grow with demand, while deduplicated slugs grow more slowly.
2. Inclusion lags. There is a delay between repository creation and Hub indexing, so new plugins appear later in the market than in repositories.
3. Archives are not removals. An archived repository may remain in the index, so deduplicated counts do not drop immediately, creating the illusion of active supply.

Do not judge by a single number. Use repository entries for ecosystem size, deduplicated slugs for choice space, and market search results for what is installable now. Read all three together.

What the numbers are good for

The total will not choose a plugin for you, but duplicate density, category distribution, and update recency will. Use these three dimensions instead of memorizing a total.

1. Check duplicate density first. A name with dozens of repositories means high duplication and large differences; inspect owner and implementation. A name with one or two repositories is easier to evaluate.
2. Then check the category center of gravity. If your need is in tools, candidates are many but similar; in memory or workflow, candidates are few but require careful comparison.
3. Finally check update recency. A commit within three months is more likely to support the current version; long-dormant projects should be assumed to need work.
4. Cross-check with the market. In repositories but not in the market suggests indexing lag; in the market but repository unreachable suggests a broken source.

Numbers narrow the range; they do not conclude. 19,783 means a large ecosystem, 16,079 means duplicate names remain common at the name level, and 1,705 duplicated names mean source verification is mandatory. The rest is about the specific plugin.

How to recompute the three numbers

All three can be recomputed from the plugin index. The method exists so conclusions are verifiable, not so you memorize one day's snapshot.

1. Take every directory containing plugin-meta.json as one repository entry.
2. Group by the last path segment (plugin slug); the number of groups is the deduplicated slug count.
3. Count owners per group; groups with more than one owner are duplicated names, and the sum of (count minus one) is the extra entry count.
4. Group by the first-level directory for category distribution; include both flat and three-level directories or you will miss part of the index.

Snapshots expire; the method does not. Replace the date when you recompute. If the numbers move sharply, check whether the index changed its directory layout before drawing conclusions.

Next steps

Replace "how many" with "which one". To see what the market actually returns, open DSH Plugin Hub. To distinguish same-name plugins, read dsh plugins vs dsh plugin. To pick a first set by scenario, read Where to install deepseek harness plugins.

Sources: DeepSeek Harness docs - Your first plugin, dshplugin/dsh-plugin-hub, GitHub topic: dsh-plugin, DSH Plugin Hub plugin page