Repo Prompt
Repo Prompt is a free, open-source macOS app for context engineering. It curates exactly what code an AI model sees, generates token-efficient codemaps, and orchestrates coding agents through a built-in MCP server or the command line.
What is Repo Prompt?
Repo Prompt is a native macOS application built with Swift and SwiftUI that concentrates on a single problem: deciding what a language model actually sees before it starts reasoning. Its premise, stated plainly on the site, is that coding agents waste their context window before they think. The publisher breaks a typical agent context down as roughly 16% system prompts and tool definitions, 38% duplicate reads and unrelated code, 16% genuinely relevant code, 14% actual reasoning, and 16% context the agent never found. The answer proposed is to assemble that context deliberately and up front: a file tree for orientation, codemaps for the API surface, complete contents for the files that matter, git diffs for change awareness, and the task description itself. The application offers three ways of working. Agent Mode hands context engineering to CLI agents such as Codex, Claude Code, Cursor, Gemini and OpenCode. A built-in MCP server exposes more than fifteen token-efficient tools to any MCP-compatible client. IDE Mode, the original manual workflow, survives only in the discontinued Classic Edition. Its central mechanism is the codemap. Tree-sitter extracts function signatures, class and type definitions, interfaces, constants and enums without implementation bodies, which the publisher reports as roughly ten times fewer tokens. Twelve languages are covered: Swift, JavaScript, TypeScript, Python, C, C++, C#, Rust, Go, Java, Dart and PHP. Around it sit Context Builder, a three-stage explore, curate and handoff pipeline working to an explicit token budget, and five prepared workflows invoked as slash commands. Everything stays local. The MCP server talks over a UNIX socket with no network port open, and API keys live in the macOS Keychain. Models are brought by the user: nine direct API providers alongside OpenRouter, local Ollama, free GitHub Models and any OpenAI-compatible endpoint, or an existing Claude, ChatGPT, Cursor or Google subscription reused through CLI Providers. Since June 2026 the project has been open source, distributed as the Community Edition on GitHub.
What it does
- Select precisely which files, line slices or signatures from a repository reach the model
- Generate tree-sitter codemaps that expose the shape of your code without its implementation
- Count and budget tokens before anything is sent
- Run Context Builder to explore a codebase, curate files within a budget and hand off a structured prompt
- Launch ready-made workflows as slash commands for orchestration, building, review, investigation and refactoring
- Orchestrate sub-agents in parallel or in sequence, steer them mid-run and verify their output
- Review proposed edits file by file as diffs, apply them selectively and undo when needed
When to use Repo Prompt / When not to
A quick filter to help you decide if Repo Prompt is the right fit.
When to use Repo Prompt
- Mac developers who work alongside coding agents and want to decide exactly which files, slices and signatures the model receives
- Engineers shipping production code, who need a plan that fits the patterns already in the repository rather than generic AI advice
- Teams that take code review seriously and want feedback grounded in real git diffs and repository context through the /rp-review workflow
- Anyone already paying for Claude Pro or Max, ChatGPT Plus or Pro, Cursor Pro or Business, or Google AI, who can reuse that subscription with no extra API spend
- People working on large repositories where the token budget is the real constraint, and who need codemaps, line slices and an explicit budget to stay inside it
When not to use Repo Prompt
- Windows and Linux users: the app is written in Swift and SwiftUI and runs only on macOS 14.0 Sonoma or later, with a waiting list standing in for a port
- Anyone expecting a web or mobile product; there is nothing to open in a browser or on a phone
- Developers looking for a full editor or IDE, since Repo Prompt feeds other tools rather than replacing Xcode or VS Code
- Users who want models included in the price, as the app supplies none and requires your own API keys or an existing CLI subscription
- Organisations needing contractual support, accounts, shared workspaces or central administration, none of which exist here
How to use Repo Prompt
A typical end-to-end flow, from setup to results.
- Download the latest .dmg from the Community Edition releases on GitHub, or install the prebuilt binaries with Homebrew
- Drag Repo Prompt CE into your Applications folder, eject the disk image, and open it from Applications or Spotlight
- Check you meet the requirements first: macOS 14.0 Sonoma or later, an Apple Silicon or Intel Mac, and about 50 MB of disk space
- Grant folder access on first launch, which the app needs to read your code, then open a workspace with File then Open or by dragging a folder onto the window
- Click two or three related files in the tree to add them to the context, watching the token counter as you go
- Right-click any file to choose between Full, Slices and Codemap modes; a plus marker means a codemap is available
- Write your task in the Instructions area, being specific about the goal, the constraints and the expected output
- Either press Copy and paste into ChatGPT or Claude.ai, or start a New AI Chat where the context is injected automatically
- Add your API keys under Settings, API Settings, API Providers, then Validate and Save; keys are stored in the macOS Keychain
- Review the returned changes as diffs in the Review view, apply all or a selection, and use Undo if a change is wrong
Pros & Cons
Pros
- Free and open source, with no account, no subscription and no product licence key required
- Your code and files are neither collected nor stored by the publisher, and processing happens locally on your machine
- The MCP server uses a local UNIX socket and opens no network port, which suits locked-down corporate networks
- Measurable token savings through codemaps reported at ten times fewer tokens, line slices and explicit budgets
- Existing Claude, ChatGPT, Cursor or Google subscriptions can be reused through CLI Providers with no additional API cost
- A wide provider choice, from nine direct APIs to local Ollama and any OpenAI-compatible endpoint
- Changes stay under human review, with file-by-file diffs, selective apply, undo and checkpoints
Cons
- macOS only, and only from version 14.0 Sonoma; Windows and Linux are left to a waiting list and to community effort
- No models are included, so an API key or a third-party CLI subscription is a prerequisite and an external cost
- Support and maintenance are explicitly best-effort, with the publisher reserving the right to limit support for versions over a year old
- The software is provided as is, without warranty, and compatibility with every macOS version is not guaranteed
- Features may be altered, improved or removed without prior notice
- There is a genuine learning curve, as codemaps, slices, presets, token budgets, workflows and MCP transports form a vocabulary to absorb
- The Classic Edition is abandoned, so anyone attached to the manual IDE Mode must migrate or stay on an unmaintained build
Pricing & Plans
Repo Prompt Community Edition is free of charge. The terms and conditions state that no purchase, paid subscription or product licence key is required to use it, and the software is distributed as open source under the licence included with its repository. The site publishes no pricing page at all, so there is no entry price point and no currency to quote. The only expenditure is external: the API keys or the existing AI subscriptions used to supply the models, which are billed by those providers and not by Repo Prompt.
- Community Edition — free and open source
- under active development
- downloaded from the project's GitHub releases. Built for modern coding agents
- with Context Builder
- codemaps
- reviewable handoffs
- MCP and CLI agent integrations
- and community-driven development.
- Classic Edition — free
- but no longer supported. The original manual workflow for selecting files
- assembling prompts and copying them into an AI tool
- with archived source available for reference and no future updates or support.
Data, GDPR & hosting
A consolidated view of how Repo Prompt handles your data.
GDPR overview
GDPR compliance is claimed explicitly. The privacy policy, effective 12 April 2025, says it is designed to comply with the EU General Data Protection Regulation and Quebec's Law 25, and covers users worldwide including the EEA. The stated legal basis is legitimate interest, limited to keeping the application stable. PostHog and Sentry are named as processors under Data Processing Agreements and standard contractual clauses, with Sentry additionally relying on the EU-U.S. Data Privacy Framework. Access and erasure rights are openly described as limited in scope, because no personal identifier is collected and no data can be traced back to an individual; opting out means blocking network access to those services. No Article 27 EU representative is named. Requests go to contact@repoprompt.com.
Who owns the data?
Your code stays yours and stays put. Repo Prompt runs locally, has no accounts and no sign-up, and its privacy policy states that it does not collect or store the contents of your code or files, your name, email, IP address, device identifiers, cookies or session data. API keys sit in the macOS Keychain, encrypted at rest, and never leave the machine. The only data reaching the publisher is anonymous usage analytics and crash reports. The named data controller is Eric Provencher, founder and privacy officer, reachable at contact@repoprompt.com. When you use optional API integrations, data travels straight from your device to that provider.
Reuse rights
Reuse rights are unusually generous, with one clear boundary. The source code is free and open source: the terms state it may be used, copied, modified and distributed under the licence included with the applicable repository. Prompt Content shipped with the software, meaning its prompts and skills, may be used and modified freely in your own projects, and files installed into your project directories may be committed and pushed to your repositories as part of normal work. What you may not do is redistribute, rehost or republish that Prompt Content as a standalone resource, bundle it into unrelated commercial products, or place it in prompt libraries and similar collections; public repositories containing it are asked to credit Repo Prompt. Brand assets, trademarks, product names and logos are excluded from the open-source licence. Your own code and files are never claimed by the publisher, which does not collect them.
Data retention & training
Hosting summary
Almost nothing is hosted anywhere but your own Mac. Repo Prompt runs locally and operates primarily on your device, and its privacy policy states that the contents of your code and files are never collected or stored. API keys are kept in the macOS Keychain, encrypted at rest, and never leave the machine, while the MCP server communicates over a local UNIX socket with no network port exposed. The only data that does travel is anonymous usage analytics, handled by PostHog, and crash reports, handled by Sentry. Both are stored in the United States. For those transfers the publisher relies on Data Processing Agreements and standard contractual clauses, and additionally on Sentry's participation in the EU-U.S. Data Privacy Framework. Optional API integrations are a separate matter: that data goes straight from your device to the model provider you have chosen, and the publisher states it does not intercept, log or transmit it.
Things to keep in mind
Risks and trade-offs to weigh before adopting Repo Prompt.
- No incorporated company is published anywhere on the site: the terms name the publisher simply as RepoPrompt, with no legal form, and the named data controller is a private individual. Check who you are actually contracting with before relying on the tool commercially.
- Support and maintenance are explicitly best-effort with no contractual commitment, and the publisher reserves the right to limit support for versions more than a year old.
- The software is provided as is, with no warranty and no liability for data loss or downtime, and features may be altered or removed without prior notice.
- Data sent to third-party APIs leaves the publisher's perimeter entirely. It is on you to read the terms of whichever model provider you connect, since Repo Prompt makes no representation about what they do with it.
- Analytics and crash reports travel to the United States through PostHog and Sentry, and the only documented way to opt out is to block network access to those services rather than to flip a setting inside the app.
- Prompt Content carries real restrictions: free to use and modify inside your own projects, but not to redistribute, rehost or republish as a standalone resource, and brand assets and logos sit outside the open-source licence.
- A tool that makes it easy to hand large amounts of context to a model can quietly encourage you to stop reading your own codebase. The diff review, selective apply and undo controls exist for a reason, and skipping them defeats the purpose of curating context in the first place.
Setup & Integrations
Technical difficulty
Getting started is easy: download the .dmg from GitHub or install through Homebrew, drag the app into Applications, grant folder access, and open a project. The site puts a first prompt at roughly two minutes. Beyond that the difficulty rises. You must obtain and validate an API key or install an agent CLI, and rp-cli needs a symlink in /usr/local/bin that prompts for administrator access. MCP setup is a one-click install for Cursor, Claude Code and Codex CLI. Migrating from the Classic Edition is the fiddliest step, requiring a data migration and corrected MCP paths.
Deployment
Integrations
Behind Repo Prompt
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
How much does Repo Prompt cost?
Which platforms does it run on?
Where do I download it?
Do I need to create an account?
Is my source code sent to the publisher?
Is my code used to train AI models?
Which AI models can I use?
Can I reuse a subscription I already pay for?
Does the MCP server open a network port?
Which law applies, and is there a minimum age?
Should you pick Repo Prompt?
Repo Prompt is an unusually focused tool, and that focus is its strength. It does one job, on one platform: making sure a language model receives the right slice of a codebase before it starts reasoning, rather than discovering the repository one wasteful tool call at a time. The codemap mechanism, the explicit token budgets and the three-stage Context Builder are concrete answers to a real and expensive problem, and the reported tenfold token saving is the kind of claim a user can verify within an afternoon. The privacy posture is genuinely strong and worth stating plainly. There are no accounts, the application runs locally, API keys stay in the macOS Keychain, the MCP server opens no network port, and the publisher states that the contents of your code and files are never collected. For anyone whose employer takes a dim view of source code leaving the machine, this matters more than any feature. Since June 2026 the project has been open source, and the shift shows in what you get and what you do not. There is no price and no licence key, but also no contractual support: maintenance is best-effort, the software is provided as is, and features may change without notice. The Classic Edition has been retired outright. Two reservations deserve weight. Repo Prompt is macOS-only, with Windows and Linux reduced to a waiting list, and it supplies no models of its own, so an API key or an existing AI subscription remains a prerequisite. There is also a real vocabulary to learn before the tool repays the effort. For a Mac developer already working with coding agents on a substantial repository, it is a strong and inexpensive addition. For anyone else, the platform requirement settles the question first.
- Choosing a selection results in a full page refresh.
- Opens in a new window.