Describe what you’re building. Get popular, free-to-use, maintained repos that fit.
Nothing you type is stored. · · · · · A Cardzo product
Repository data from the GitHub API. Replore is not affiliated with or endorsed by GitHub. GitHub® is a trademark of GitHub, Inc.
A build plan (a PRD, or product requirements doc) your AI can code straight from. Replore doesn't write it: your own AI does, free and private. Copy the prompt below into any AI agent (Claude Code, Codex, Cursor…) or open it in Claude, ChatGPT or Gemini. It lists the repos you selected.
Prompt copied. Paste it into Gemini (it has no pre-fill link).
Long plan: each button copies the prompt first, then opens the assistant. Paste it there.
Got a useful plan? Leave a tip It helps us keep Replore free and fund more open-source projects.
Plans work best with 10 or fewer repos. You picked . Untick the weakest few for a sharper plan.
Open in Claude opens a new chat on your own claude.ai account with the prompt ready to send. (If it doesn't pre-fill, tap the copy icon and paste.) New to the plugin? .
Narrow your search. All optional. These apply to your next search.
Paste a repo you already like (owner/repo or a GitHub link), then type a few words in the
search box (e.g. “free ai call”) so I can narrow the similar repos to what you actually want.
Three steps, right in your browser. No account, no key.
Describe your project, tap Find, and get the repos that fit: relevant, popular, free to use.
Hybrid search: repos are ranked by your words and by what you mean, so "voice notes" also finds "speech-to-text". Repos our index finds that the word search missed are checked on GitHub before you see them.
Build a plan: tick the repos you like on the results, then tap Generate plan. Replore hands the picks to your own Claude (Copy, or Open in Claude), which writes a full build plan (a PRD) you can code straight from, free and private.
For AI-written searches and a whole PRD read feature by feature, use the plugin (Claude Code, Codex, Cursor or any AI agent).
Your own AI writes the searches, judges every result and shows every repo that fits, not just this page's list. Free and open source. Pick your agent:
The full plugin: your AI writes the searches, judges every result and reads a whole PRD feature by feature.
Connect Replore through MCP, or just open the cloned folder: Codex reads AGENTS.md on its own.
Add Replore to Cursor's MCP settings, then ask for repos in the chat.
Add Replore as an MCP server, then ask Gemini for repos that fit your idea.
Add Replore to Copilot's MCP servers. Copilot also reads AGENTS.md in the folder.
Add Replore to Windsurf's MCP settings, then ask for repos in the chat.
Add Replore in Cline's MCP servers panel, then ask for repos for your project.
Command python3, argument scripts/mcp_server.py. Works anywhere MCP (Model Context Protocol) does.
Open the cloned folder and say "follow AGENTS.md". Your agent runs Replore's search (scripts/scout.py) itself.
python3 --version (Windows: python.org, tick "Add python.exe to PATH"; Mac: brew install python).claude in any folder, then paste:/plugin marketplace add chittychat-studio/replore /plugin install replore@replore
/replore:find a diet app for iPhone, starting from scratch
python wherever it says python3. Then cd replore and run pwd (Windows: cd): that prints the full path you need next.git clone https://github.com/chittychat-studio/replore
python3, argument <that path>/scripts/mcp_server.py. Add the replore entry inside the existing list; don't replace the file. Then restart Codex. The exact file and snippet for each agent are in the README.find_repos tool and shows the comparison table.python wherever it says python3. Then cd replore and run pwd (Windows: cd): that prints the full path you need next.git clone https://github.com/chittychat-studio/replore
python3, argument <that path>/scripts/mcp_server.py. Add the replore entry inside the existing list; don't replace the file. Then restart Cursor. The exact file and snippet for each agent are in the README.find_repos tool and shows the comparison table.python wherever it says python3. Then cd replore and run pwd (Windows: cd): that prints the full path you need next.git clone https://github.com/chittychat-studio/replore
python3, argument <that path>/scripts/mcp_server.py. Add the replore entry inside the existing list; don't replace the file. Then restart Gemini CLI. The exact file and snippet for each agent are in the README.find_repos tool and shows the comparison table.python wherever it says python3. Then cd replore and run pwd (Windows: cd): that prints the full path you need next.git clone https://github.com/chittychat-studio/replore
python3, argument <that path>/scripts/mcp_server.py. Add the replore entry inside the existing list; don't replace the file. Then restart VS Code Copilot. The exact file and snippet for each agent are in the README.find_repos tool and shows the comparison table.python wherever it says python3. Then cd replore and run pwd (Windows: cd): that prints the full path you need next.git clone https://github.com/chittychat-studio/replore
python3, argument <that path>/scripts/mcp_server.py. Add the replore entry inside the existing list; don't replace the file. Then restart Windsurf. The exact file and snippet for each agent are in the README.find_repos tool and shows the comparison table.python wherever it says python3. Then cd replore and run pwd (Windows: cd): that prints the full path you need next.git clone https://github.com/chittychat-studio/replore
python3, argument <that path>/scripts/mcp_server.py. Add the replore entry inside the existing list; don't replace the file. Then restart Cline. The exact file and snippet for each agent are in the README.find_repos tool and shows the comparison table.python wherever it says python3. Then cd replore and run pwd (Windows: cd): that prints the full path you need next.git clone https://github.com/chittychat-studio/replore
python3, argument <that path>/scripts/mcp_server.py. Add the replore entry inside the existing list; don't replace the file. Then restart Any MCP agent. The exact file and snippet for each agent are in the README.find_repos tool and shows the comparison table.git clone https://github.com/chittychat-studio/replore
AGENTS.md automatically; with any other agent say "follow AGENTS.md". It then runs scripts/scout.py itself.Stuck? Full guide: github.com/chittychat-studio/replore.
Three steps from a shortlist to building.
Remember: stars are popularity, not quality, and a licence verdict is a plain summary, not legal advice.
Hi, it's Leo! Found a bug, or want something new? Tell me and I'll pass it straight to the Cardzo team (hello@cardzo.co.in).
Tell me what it's about first.
Got it, thanks! I passed it to the Cardzo team.
Or write to hello@cardzo.co.in
Replore has no account and no user database, and stores nothing you type. Our host counts page views with Cloudflare Web Analytics, without cookies (see below).
static.cloudflareinsights.com. It records the page
address (your searches are never in it), the referring site, your browser, operating system and device type, your
country and how fast the page loaded. It sets no cookie, stores nothing on your device, never sees what you type and
gives us only totals. Cloudflare handles it under its
privacy policy.api.web3forms.com), an email-forwarding service, to hello@cardzo.co.in. Nothing is
sent until you press Send.The plugin and MCP server run on your computer; your AI provider handles your conversation under its own terms. Full details, including the plugin's local safety log: PRIVACY.md. This is a plain description of how Replore works, not legal advice.
Results are ranked by fit: how well a repo matches your words, how popular it is (stars) and how freely you can use its licence. So the top pick is relevant, popular and free to use, not just the most-starred.
How Leo searches: he splits what you describe into separate needs and searches GitHub for each, specific needs first. An AI check then reads the top 30 and moves repos that don't fit to the end, under "Less likely to fit". If the check is unavailable you get the keyword results unchanged. A search you repeat within an hour comes back instantly from your own browser.
On each card: when the code last changed, the licence in plain words (free for commercial use or not), stars, the language and whether it's a starter template. The filters on the left and the sort menu only narrow and reorder what was found; they never search again. Details shows why it fits, Diagram, Wiki and Code to look inside, and Is it safe to build on? (OpenSSF Scorecard, known vulnerabilities, releases), which is for information and never changes the ranking.
The goes further: your own AI writes the searches, judges every result and reads a whole plan feature by feature.
Limits: results are GitHub search matches, not a scan of every repository. Stars measure popularity, not quality, and can be bought. The licence verdict comes from the licence only: a plain summary, not legal advice. Some projects also sell a paid hosted version or add-ons.
Hybrid search: repos are ranked by your words and by what you mean, so "voice notes" also finds "speech-to-text". Repos our index finds that the word search missed are checked on GitHub before you see them.