Requirements for Effective Use

To get reliable, high-quality results from Generacy (or any AI development agent), your project must meet several baseline requirements. These are the same fundamentals a human developer needs: access, tools, and understanding.

💡 Already using AI tooling?

If your project already uses Claude Code or another MCP-compatible agent, most of these requirements are likely satisfied. Skim the checklist at the bottom to confirm.

1. Execution Access

The agent must be able to run and test your software locally.

This is the single most important requirement. Without it, the agent cannot verify its own work, debugging becomes guesswork, and output quality drops significantly.

Your project needs:

A working local development environment (builds successfully, dependencies resolved)
A clear command to start the application (e.g., npm run dev, docker compose up)
The ability to execute code changes and observe results

⚠️ Web Applications

For web applications, browser automation (e.g., Playwright) is additionally required so the agent can interact with the UI, verify visual changes, and run end-to-end tests.

2. Tool Access (MCP Servers)

The agent must have access to the tools needed to interact with your project.

These tools are provided via MCP (Model Context Protocol) servers. At minimum, the agent needs:

Capability	Purpose
File system access	Read and write project files
Terminal / CLI	Run commands, build, and test
Browser automation	Interact with web UIs (for web apps)

Relevant MCP servers must be installed, configured, and functional. The specific servers depend on your project type — a CLI tool needs terminal access, while a web app additionally needs browser control. Without working MCP servers, the agent cannot meaningfully interact with your environment.

3. Project Context

The agent must be given structured context about your project.

This is typically provided via a CLAUDE.md file (or equivalent context file) at the root of your repository. It should include:

Project purpose — what the project does and why
Tech stack — frameworks, languages, and key dependencies
How to run locally — commands to build, start, and test
Architecture overview — key directories, services, and how they connect
Testing approach — how to run tests and what testing tools are used

Without structured context, the agent must infer everything from the codebase alone. This degrades both accuracy and speed significantly.

💡 Getting started with a context file

If your project doesn't have a CLAUDE.md yet, start with the basics: what the project is, how to run it, and where the important code lives. You can iterate from there. See the Onboarding Guide for the full project setup workflow.

4. Iterative Feedback Loop

The environment must support a generate → run → test → fix cycle.

The agent works best when it can:

Make a change
Run the application or tests
Observe the result
Fix any issues

This requires fast, reliable execution and the ability to run tests or validations programmatically. Without this feedback loop, the agent cannot improve its own outputs, and progress becomes manual and slow.

5. Onboarding New Projects

If your project has not previously used AI tooling, you will need to:

Create a context file (e.g., CLAUDE.md) with the information described above
Set up MCP servers appropriate for your project type
Verify local execution works reliably before engaging the agent

💡 Already set up?

If the project already uses AI tooling (e.g., an existing Claude Code or Cursor setup), these requirements are likely already met. Check the checklist below to confirm.

Summary Checklist

Use this checklist to verify your project is ready:

Local execution works reliably (build, start, test)
Agent can run and test code changes
Playwright (or equivalent) is available for web applications
MCP servers are installed and functional
A CLAUDE.md (or equivalent) provides clear project context
The feedback loop (generate → run → test → fix) is operational

Bottom line

Generacy requires the same fundamentals as a human developer — access, tools, and understanding. Without those, performance degrades quickly regardless of prompting. These are not Generacy-specific optimizations; they are baseline requirements for any AI development agent.

↑ Back to top