Give your teams and agents the whole story

CoreStory reads every line of your code and maps how your business actually works, from applications and processes down to business rules, each traced to the exact code behind it. Your teams and every AI agent you use work from the same understanding.

See how to get set up ↓

Start with your own code today

Desktop app for Windows coming soon.

Linux AppImage and the command-line tool for every platform are on the latest release page.

Download for free with your own model key, point it at a repo, and connect your coding agent.

Or tell your coding agent:

Install CoreStory from get.corestory.ai and set it up for this repository.

Your agent installs the command-line tool and connects itself. You accept the terms and add your model key when it asks you to run corestory setup.

Works with GitHub Copilot, Claude Code, Cursor, Codex and any other agent that supports MCP.

What your agents miss without the full story

Broken business rules

Business rules are often spread across several files and services, and your agent only sees the ones in front of it. When one is changed, it misses the rest. Defects can reach customers and your team spends weeks on rework.

Answers you can’t check

Your agent explains how the system works with full confidence, but it is guessing from the few files it read. Your team can’t tell which parts are true without digging through the code themselves.

Lost features in rewrites

Old code often handles edge cases that nobody documented. When your agent rewrites it without knowing they exist, features your customers rely on can disappear after launch.

Wasted AI spend

Without a map of the codebase, your agent tries one fix after another in the wrong places. Your token bill climbs, and an engineer still has to step in and finish the job.

The difference the right context makes

We tested four frontier models on hard enterprise-style code changes, with and without CoreStory. Every one of them did far better with the full picture.

40–74%

of hard code changes that agents alone usually fail were fixed or flagged, across every frontier model we tested

98%

of the places business rules live in the code were covered

17–35%

lower AI spend for each business rule found

Read the full benchmark

Get set up

You don’t need a sales call to get started, and the free desktop app runs on your own machine.

Download and open

Install the desktop app, accept the terms, and add your OpenAI, Azure OpenAI or Anthropic key. No account needed.

Point it at a repo

Pick a repository folder. CoreStory indexes it on your machine and builds a map of how the system connects and behaves.

Plug in your coding agent

Connect your agent over MCP with the corestory command-line tool so it can use the map while planning and editing.

System requirements

Supported platforms and what you need before you install.

See requirements ↓

Install and add your model key

Step-by-step install, plus which model providers are supported and where the key is stored.

Install steps ↓

Connect your coding agent

Setup for Claude Code, Codex, GitHub Copilot and Cursor.

Connect your agent ↓

Agent Skills

26 ready-made skills that teach your agent how to use CoreStory, for both hosted and local setups.

View the skills on GitHub

Release notes and checksums

What changed in each version, plus checksums to verify your download.

Latest release

Guides

Documentation and getting-started guides for CoreStory.

Read the docs

System requirements

  • macOS: macOS 12 Monterey or later on a Mac with Apple silicon. Desktop app and command-line tool.
  • Linux: x64, as a .deb package or an AppImage, plus the command-line tool. The desktop app needs WebKitGTK 4.1 (for example, Ubuntu 22.04 or later).
  • Windows: Windows 10 or later, x64. Command-line tool only for now; the desktop app for Windows is coming soon.
  • A model provider key from OpenAI, Azure OpenAI or Anthropic.
  • A repository on your machine to point CoreStory at.

Install and add your model key

Desktop app

  1. Download the app for your platform from the download buttons above and open it.
  2. Agree to the terms of use and privacy policy. A checkbox lets you choose whether to send anonymous usage data.
  3. Choose your model provider (OpenAI, Azure OpenAI or Anthropic) and paste your key. CoreStory saves it to ~/.config/corestory/config.toml, readable only by you.
  4. Pick a Git repository folder. CoreStory indexes it locally and opens the map.

You only see the terms and model steps the first time. After that, the app opens straight to the folder picker.

Command-line tool

Coding agents: follow the step-by-step install instructions at get.corestory.ai/llms.txt.

Download the command-line tool for your platform from the latest release and put corestory on your PATH. Run setup once to accept the terms and add your model key, then index a repository:

corestory --version
corestory setup
cd your-repo
corestory init
corestory ingest

CoreStory keeps its index in a .corestory folder inside the repository. Add .corestory/ to your .gitignore.

Connect your coding agent

Your agent talks to CoreStory through the corestory command-line tool, which serves the repository’s map over MCP. Install the command-line tool and run corestory setup first (see above). CoreStory serves the repository it is started in, so each setup below starts it from your repository folder.

Claude Code

Install the local CoreStory plugin. It connects CoreStory to the repository you open and adds the CoreStory skills. It needs no account and makes no network calls.

claude plugin marketplace add corestoryai/agent-skills
claude plugin install corestory-local@corestory

GitHub Copilot (VS Code and CLI)

Add a .mcp.json file at the root of your repository. Commit it to share the setup with your team. Claude Code reads this file too, if you would rather not use the plugin.

{
  "mcpServers": {
    "corestory": {
      "type": "stdio",
      "command": "corestory",
      "args": ["mcp"]
    }
  }
}

Open the repository folder in your agent. Copilot CLI asks you to trust the folder, and Claude Code asks you to approve the server the first time.

Cursor

Add .cursor/mcp.json to your repository. Cursor has no working-directory setting, so this starts CoreStory from the project folder (macOS and Linux):

{
  "mcpServers": {
    "corestory": {
      "type": "stdio",
      "command": "sh",
      "args": ["-c", "cd \"$0\" && exec corestory mcp", "${workspaceFolder}"]
    }
  }
}

Codex

Add the server to ~/.codex/config.toml, with cwd set to your repository:

[mcp_servers.corestory]
command = "corestory"
args = ["mcp"]
cwd = "/absolute/path/to/your-repo"

Agent Skills

Teach your agent CoreStory’s workflows with the Agent Skills. For Codex, install them with the GitHub CLI:

gh skill install corestoryai/agent-skills --all --agent codex --scope user

Using hosted CoreStory with your organization? Follow the Agent Skills install guide with your organization’s MCP URL.

Every system has a story and now everyone can read it

Download CoreStory for free and give your engineers and agents the same understanding of your business.

Get it free

Rolling this out across a large team? Talk to an expert