#What is AgentCN?

AgentCN is a shadcn-style registry for installing complete backend AI agent recipes rather than visual UI components. Each recipe includes the agent definition, instructions, tools, skills, workflows, and supporting files required to add a working agent to an existing Eve or Flue project. It is designed for developers and product teams building AI research tools, document assistants, coding agents, internal knowledge systems, workflow automation, Slack bots, database assistants, and browser-based agents without relying on restrictive black-box packages.

#Features ⚡

• Complete Agent Recipes: Every registry item can include instructions, tools, skills, workflows, agent definitions, and supporting utilities instead of only a basic prompt.

• Shadcn CLI Installation: Agents can be installed directly with the familiar shadcn add workflow using the AgentCN registry namespace.

• Full Source-Code Ownership: Recipe files are copied into the developer’s project, allowing every model, instruction, tool, schema, and workflow to be modified.

• Eve and Flue Support: Recipes are provided for Eve, a filesystem-first Vercel agent framework, and Flue, an agent runtime supporting subagents, skills, and bounded workflows.

• Live Agent Previews: Developers can run supported agents directly from their documentation pages before installing them.

• MCP Integration: AgentCN recipes can be discovered and installed from compatible development environments through the shadcn MCP server.

• Model Customization: Developers can replace the default AI model with another supported model by editing the installed agent source.

• Composable Architecture: Agents, tools, skills, and workflows can be combined to create more advanced AI pipelines.

• TypeScript Support: Recipes use TypeScript for typed agent definitions, structured outputs, tool parameters, and validation.

• External Tool Integration: Recipes can connect with web-search providers, databases, Slack, Google Sheets, GitHub, vector stores, browsers, and document-processing services.

• AI SEO Auditing: The AI SEO Audit agent runs approximately 30 deterministic checks across six categories and generates an actionable optimization prompt.

• Retrieval-Augmented Generation: Document-focused recipes can index PDFs and internal files into vector stores and return source-grounded responses.

#Pro and cons

Pro ✅:

• AgentCN is completely free and open source under the MIT License.

• Developers receive the complete agent source instead of importing an opaque package.

• Installation follows the familiar shadcn registry workflow.

• Recipes can be customized without depending on a proprietary AgentCN runtime.

• Both Eve and Flue versions are available for the current recipe collection.

• The library covers a broad selection of practical production use cases.

• Live previews make it easier to evaluate agents before installation.

• MCP integration allows AI coding tools to discover and install recipes.

• Models, providers, tools, prompts, and output schemas can be replaced.

• Commercial use, modification, and redistribution are permitted under the license terms.

Cons ⚠️:

• AgentCN provides backend agents, not visual shadcn UI components.

• An Eve or Flue project must be configured before installing most recipes.

• Many recipes require external API keys and separately billed services.

• Developers are responsible for hosting, model usage, storage, and provider costs.

• Recipes that access databases, browsers, files, Slack, or company documents require careful permission and security reviews.

• Advanced customization requires knowledge of TypeScript, AI models, tool calling, and agent workflows.

• AgentCN is a relatively new project, with its initial release published on June 20, 2026.

• Self-hosted recipes require developers to manage deployment, observability, rate limits, and production safeguards.

#Included Components/Templates

AgentCN includes 19 production-ready agent recipes as of June 25, 2026, with Eve and Flue versions available across the current catalog.

• Deep Search: Conducts iterative web research, evaluates information gaps, and returns a cited answer.

• AI SEO Audit: Scores webpages for visibility and readability across AI answer engines.

• CSV to Questions: Summarizes CSV datasets and generates focused analytical questions.

• Customer Feedback Summary: Categorizes customer feedback and creates an executive report with recommendations.

• Meeting Notes: Converts meeting transcripts into summaries, decisions, and action items.

• Chat with PDF: Indexes PDF documents and answers questions with page-level citations.

• Flash Cards from PDF: Generates study flash cards from PDF content with optional images.

• Chat with YouTube: Answers questions using video metadata and transcripts with timestamp citations.

• Docs Chatbot: Answers questions using structured library and API documentation.

• Chat with Database: Inspects database schemas, converts natural-language questions to SQL, and executes read-only queries.

• GitHub PR Review: Produces adaptive, file-by-file feedback for GitHub pull requests.

• Slack Agent: Responds to Slack mentions and direct messages within the appropriate thread.

• Google Sheets Agent: Reads, analyzes, and updates Google Sheets through the Sheets API.

• Weather Agent: Retrieves current weather information through the Open-Meteo API.

• Docs Expert: Searches current online documentation and returns source-cited answers.

• Browser Agent: Controls a browser through Playwright using snapshots and selectors.

• Claw Assistant: Works inside a sandboxed environment to read files, write files, and execute commands.

• Company Knowledge: Indexes internal documents and provides answers with personally identifiable information redaction.

#Pricing 💸

AgentCN follows a free and open-source pricing model. There is no publicly listed Pro, premium, subscription, or one-time purchase tier as of July 1, 2026.

• Agent Recipes: Free

• Source Code: Free and publicly available

• License: MIT License

• Commercial Use: Permitted

• Modification and Redistribution: Permitted when license requirements are retained

• Pro Tier: No paid AgentCN tier is currently listed

• Optional Support: Users can sponsor the project through its available sponsorship links

• External Costs: AI models, hosting, databases, vector stores, Bright Data and other third-party services may charge separately

For example, the Deep Search recipe uses the Bright Data SERP API by default, while some previews and recipes require Anthropic or API keys. These provider charges are separate from AgentCN itself.

#Integrations 🧰

• Shadcn CLI and shadcn registries

• Shadcn MCP server

• Eve

• Flue

• TypeScript

• Vercel Functions

• Next.js

• Anthropic and Claude models

• Bright Data SERP API

• Bright Data Browser and Web Unlocker

• GitHub

• Slack Web API

• Google Sheets API

• Playwright

• Open-Meteo API

• SQL databases

• libSQL vector storage

• Pinecone, Qdrant, Chroma, and pgvector through customization

• PDF processing and embedding models

• MCP-compatible coding editors

• npm, pnpm, Yarn, and Bun workflows

Individual integrations depend on the selected recipe. AgentCN documents the required packages and environment variables on each agent page.