Direct answer

Agent Skills Directory

Agent Skills Hub is a screening directory for finding MCP servers and reusable agent skills by task, source, permission risk, and recorded review signals. The current dataset contains 1,213 entries across 20 categories. Use the filters to build a shortlist, then open the listing and verify its source before installation; a directory grade is decision support, not a security guarantee.

Directory entries: 1,213
Reviewed labels: 1,005
Official entries: 120
Latest recorded scan: 2026-05-20

Scan dates are dataset snapshots. Recheck repository activity, permissions, releases, and install instructions at the source before production use.

The most common failure in agent tooling adoption is not feature quality. It is mismatch between tool behavior and team constraints. Some teams need strict auditability, some need rapid prototyping, and others need lightweight integrations that can run in managed environments. We surface context that helps you choose the right depth of integration for your own process rather than copying somebody else's stack.

AI & ML AI & Machine Learning API & Integration Automation & Workflow Communication Data & Analytics DevOps & Infrastructure Development

How to use this directory effectively

First, filter by primary task: coding, data workflows, automation, content operations, or support. Then compare installation complexity and expected permissions. If two skills look similar, prioritize the one with clearer documentation, active issue handling, and transparent release notes. For production use, run a staged rollout: pilot in a sandbox, record failure modes, and only then move to a shared team environment.

We also recommend documenting your own acceptance checklist before installing any third-party skill. Include required capabilities, forbidden behaviors, and fallback plans if the skill becomes unmaintained. That single step reduces operational risk dramatically because your decision process becomes repeatable instead of ad hoc.

This page combines searchable listings with original decision guidance, transparent review labels, and clear usage boundaries. That context is more useful than a raw link list because teams can compare workflow fit and operational risk before opening a repository or running an install command.

Exact-match intake

Where to send package-name and workflow searches first

Fast directories can rank by raw listing count, but operators need a safer first click. Use these routes when a search looks like an install command, package name, or MCP workflow shortcut.

Database MCP setup

Use when a query includes Neon, Postgres branching, schema inspection, or an npx setup command.

Interactive MCP Apps

Use when the user is looking for MCP interfaces, UI components, or app-like server workflows.

Desktop extensions

Use when a query mentions DXT, MCPB, Claude Desktop extensions, install bundles, or one-click setup.

Browser automation

Use when the task needs page actions, screenshots, replay evidence, scraping, or UI regression checks.

Evaluation checklist before installation

Define one target workflow and one measurable success metric.
Verify required permissions against your internal policy boundaries.
Review maintenance signals: release cadence, issue response, and documentation depth.
Run a bounded pilot with log capture and rollback readiness.
Promote only after acceptance criteria pass with reproducible evidence.

Teams that follow a checklist like this avoid most expensive adoption failures. The objective is predictable workflow improvement, not collecting as many tools as possible. A narrower, measurable rollout is usually safer and faster than broad deployment with unclear ownership.

Worked example: evaluating two similar skills

Imagine you have two options that both claim to automate the same coding task. Skill A has higher stars, while Skill B has clearer install docs and lower permission requirements. In production environments, Skill B is often the better first choice because operational predictability beats popularity. Start with a one-week pilot and measure first-pass success, intervention count, and rollback frequency.

If Skill A later proves more capable without raising risk, you can switch with confidence because your baseline is already documented. This is the core advantage of structured evaluation: decisions remain reversible and evidence-based.

Prefer clearer operational boundaries over generic feature lists.
Keep rollout scope narrow until outcome quality is stable.
Use the same acceptance template for every new skill trial.

Frequently asked questions

How should teams choose the first skill to adopt?

Start with one high-frequency workflow and pick the skill that removes the biggest repeat bottleneck in that workflow.

Is a higher star count always better?

Not always. Maintenance quality, permission scope, and integration clarity usually matter more than raw popularity.

What is the safest rollout pattern?

Run a preview pilot, measure failure modes, and promote gradually with explicit rollback rules.

How often should installed skills be reviewed?

Review monthly, and run an extra check after upstream breaking changes or major dependency updates.

Why does this page include long-form guidance?

Directory pages perform better when they provide original decision support, not only link lists or shallow summaries.

Loading directory...