Documentation

Introduction

Reagent is a community-driven registry and structured discovery layer for reusable AI agent capabilities. Its purpose is simple: help developers find, understand, and reuse the building blocks that power modern AI agent systems.

The problem Reagent solves

Modern AI agents are assembled from reusable components: prompts, workflows, skills, tool integrations, and MCP servers. Right now these components are scattered across GitHub repositories, buried in blog posts, locked in private codebases, and duplicated across teams. Developers spend significant time rebuilding capabilities that already exist and work well. Good implementations are simply hard to find.

Reagent centralises this knowledge. Every resource is documented in a consistent, structured format so it can be found by search, understood without external explanation, and integrated immediately.

What Reagent is not

• Not an execution environment — resources are not run on Reagent servers
• Not a marketplace — resources are free to discover; there are no paid plugins
• Not a code editor or hosting platform
• Not an agent framework — Reagent does not define how agents must be built
• Not a social network — there are no feeds, followers, or engagement metrics

Core Concepts

Resource

A Resource is the fundamental unit of Reagent. It represents any reusable building block that can be plugged into or used by an AI agent system. Every resource is described in a standardised format covering what it does, how it works, how to integrate it, and where to find its source.

The three user actions

1. Discover

Browse and search the registry to find capabilities that already solve your problem. Use semantic search across names and descriptions, filter by resource type (skill, workflow, MCP server, prompt pack), and filter by tags. The goal is to surface something you can use immediately.

2. Understand

Every resource detail page is designed to answer one critical question: can I plug this into my agent system right now? Each resource includes a full description, an explanation of how it works internally, an installation command, and a working usage example. You should never need to leave Reagent and read an external blog post before integrating a resource.

3. Contribute

Submit reusable capabilities you have built or found. The system values reusability over originality. A well-documented, clearly explained resource is highly valuable even if it covers a workflow that others have built before.

Resource Types

Reagent recognises four resource types. Each type has a distinct purpose in an agent system.

Choosing the right type

Use Workflow when the capability involves multiple sequential or parallel steps that must be orchestrated. Use Skill when the capability is a single atomic operation. Use MCP Server when you are exposing an external system's API to an LLM. Use Prompt Pack when the capability is expressed entirely through carefully crafted prompts and system messages.

Discovering Resources

Search

The registry search bar runs against resource names and descriptions. Type any keyword related to the capability you need. The search is case-insensitive and matches partial strings, so searching 'email' will surface email-summarizer as well as any resource with 'email' in its description.

Filtering by type

Use the Type filter in the sidebar to narrow results to a specific resource category. You can select multiple types simultaneously. For example, selecting both Workflow and MCP Server will show all resources in either category.

Filtering by tags

Tags are free-form keywords that authors attach to resources during submission. Common tag families include domain tags (code, data, text, web), integration tags (git, api, mcp), and function tags (security, productivity, reasoning, ci). Select one or more tags to filter. Results must match at least one selected tag.

Sorting

Three sort modes are available: Most popular (net vote score, highest first), Newest (submission date, most recent first), and A to Z (alphabetical by resource name).

Reading the resource card

Integrating a Resource

From the detail page

Every resource detail page has an Installation tab. It contains the exact install command and a working code example. Both have one-click copy buttons. In most cases, integrating a resource is a three-step process:

• Run the install command to add the package to your project
• Copy the usage example into your agent configuration or codebase
• Adjust the configuration options to match your environment

Install command

Resources distributed as npm packages use the Reagent CLI shorthand:

npx reagent add <resource-name>

This is equivalent to running the underlying npm install command. Resources may also specify alternative installation methods (GitHub Actions, Docker, or manual config) in their detail page.

MCP Server resources

MCP Server resources run as standalone processes. Start the server with your credentials, then point your agent framework's MCP configuration to its local port:

# Example: github-connector
GITHUB_PAT=your_token npx @reagent/github-connector --port 3100

// In your agent config:
{
  "mcpServers": {
    "github": {
      "url": "http://localhost:3100"
    }
  }
}

Prompt Pack resources

Prompt packs are imported as JavaScript/TypeScript modules and return a string you pass as your system message:

import { cotBasePrompt } from "@reagent/chain-of-thought-base";

const systemPrompt = cotBasePrompt({ variant: "reasoning" });

const response = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [
    { role: "system", content: systemPrompt },
    { role: "user", content: userQuery },
  ],
});

Contributing

Any registered user can submit a resource. Contributions are the lifeblood of the registry. The submission form captures the structured information every resource must include.

What makes a good submission

• A clear, specific name in kebab-case (e.g. pdf-parser, not my-cool-tool)
• A one-sentence description that states exactly what the resource does
• A detailed long description covering what problem it solves and for whom
• An honest explanation of how it works internally — not marketing copy
• A working install command that actually installs the resource
• A real usage example with real configuration options, not pseudocode
• Accurate tags that reflect the domain and technology of the resource
• A link to the source (GitHub repo, docs site, or specification)

Submitting an existing resource you found

You are encouraged to submit resources built by others. When doing so, attribute the original author in the author field and link to the original source. The Reagent community values discovery — surfacing a great tool that was previously buried in a GitHub repo is a genuine contribution.

Publishing model

Submitted resources go through a lightweight review before appearing publicly. The review checks that the required fields are complete, the install command resolves to a real package or endpoint, and the resource does not duplicate an existing entry without meaningful differentiation. Most submissions are reviewed within 24 hours.

Resource Specification

Every resource stored in Reagent conforms to a structured schema. This schema ensures consistency across all resources and makes them comparable, searchable, and immediately usable.

Naming conventions

Resource names must be kebab-case, all lowercase, and describe the capability (not the author or tool it wraps). Good: email-summarizer, github-connector, chain-of-thought-base. Bad: JohnsTool, my-openai-wrapper, tool-v2.

Tag conventions

Use existing tags where possible to keep the tag space clean. Common approved tags: code, data, text, web, git, api, mcp, security, productivity, reasoning, ci, docs. New tags are accepted when no existing tag covers the domain.

Authentication

What requires authentication

Creating an account

Register at /register with an email and password, or use Google OAuth for one-click signup. Account creation is free with no subscription or credit card required.

Sign in

Sign in at /login. Sessions persist across browser restarts. Use the Sign out option in your account menu to end your session on any device.

Community & Moderation

Voting

Authenticated users can cast one upvote or one downvote per resource. The net score (upvotes minus downvotes) reflects community confidence in the resource's quality and usefulness. Votes are not weighted by user reputation in the current MVP.

Reviews

Reviews consist of a 1-5 star rating and a written explanation. Good reviews describe your specific integration experience: what worked, what required configuration, what the real-world performance looked like. Reviews that simply repeat the resource description without adding new information are removed during moderation.

Moderation principles

• Resources must describe real, working, reusable capabilities
• No spam, self-promotion without substance, or duplicate submissions that add no value
• Reviews must be based on genuine integration experience
• Voting manipulation (multiple accounts, coordinated brigading) results in account suspension
• Resources that are abandoned and no longer functional may be archived after community notification

Reporting

Use the report button on any resource or review to flag content that violates community standards. Reports are reviewed by the moderation team within 48 hours.

FAQ

Is Reagent free to use?

Yes. Browsing, discovering, and integrating resources is completely free with no account required. Submitting resources and community features require a free account.

Can I submit a resource I did not build?

Yes. Surfacing great resources that are currently buried is a genuine contribution. Attribute the original author and link to the source. Do not claim authorship of something you did not build.

What if the resource I want to submit is not an npm package?

Resources do not need to be npm packages. A resource can be a GitHub Actions workflow, a Docker container, a raw prompt template, a configuration file, or any other distributable format. Use the installCommand field to describe how to get it, whether that is a shell command, a git clone, or a manual config step.

How is Reagent different from GitHub?

GitHub is a code hosting and collaboration platform. Reagent is a discovery and documentation layer. Resources listed on Reagent may live on GitHub, but Reagent's value is the structured, searchable, comparable index and the standardised documentation format — not the hosting.

How is Reagent different from npm?

npm indexes code packages. Reagent indexes agent capabilities described in human-readable form. A resource on Reagent may be distributed via npm, but it can also be a prompt, a workflow definition, or an MCP server config — none of which are npm packages. Reagent focuses on understanding and discovery, not package resolution.

Can resources be updated after submission?

Yes. The original submitter and platform moderators can update resource details. Version history is tracked so users can see what changed between versions.

What happens to low-quality resources?

Resources with consistently low vote scores and negative reviews are reviewed by moderators. If the quality issues cannot be resolved by the submitter, the resource may be archived (no longer shown in search results) but remains accessible via its direct URL for archival purposes.