Meme Radar Open Source: a read-only multi-chain token scanner you run on your own machine

The gap Meme Radar Open Source fills between a DEX screener and a contract audit

A screener tells you a token exists and is moving. A contract scanner tells you whether the token contract hides a mint function or a transfer tax. Neither answers the question a meme trader actually asks first: is this thing worth spending ten minutes on, and what is the evidence for and against it? Meme Radar Open Source is built around that intermediate step. The README describes it as a local, read-only multi-chain meme candidate radar that uses GMGN for discovery and labelling, GoPlus for contract risk review on supported chains, and DexScreener for market cap, liquidity and website cross-checking.

The intended user is someone who already does manual meme research and wants the collection work compressed. Seven chains are supported for scanning: Solana, BNB Chain, Base, Ethereum, Robinhood Chain, Arc and Stable, with one to three chains polled at a time. The output is deliberately narrow. The project states that results fall into three tiers, reject, pending review and viewable, and that viewable only means the token enters a manual research list. It does not mean the token is safe or will go up. That framing matters, because it is the opposite of how most token scanners market themselves.

The author is credited as DeFi狙击手, with an X account linked in the README. This is an open-source edition carved out of a private tool the author was already running, which explains both the scope and the omissions.

How the scan pipeline actually works: discovery, deep audit and shadow tracking

Two loops run separately, and the separation is the most interesting design decision in the project. The instant discovery window reads GMGN's one-minute turnover ranking, up to 100 records, with a target refresh of about every 20 seconds. The strict deep audit runs on its own schedule and is not driven by that heat ranking. The README is explicit that instant popularity is never used as a substitute for a security conclusion.

Deep audit pulls contract permissions, LP data, tax and honeypot risk, holder structure, ordinary wallet proxy samples, smart money, five-minute market state and price behaviour. Unknown fields are not treated as passing. GoPlus and DexScreener fill in contract risk, market cap, liquidity and website cross-checks on chains where they are supported, and missing or conflicting data is labelled as such rather than smoothed over. By default each round deep-audits at most six tokens and is bounded by a time budget. Requests go out serially by endpoint weight, and an explicit security rejection ends the audit early. Caching tops out at 60 seconds, and rate limits are respected by waiting for server cooldown rather than bypassing plan limits.

Verification is tracked as shadow performance. The passing group is followed by contract address at 5 minutes, 15 minutes, 30 minutes, 1 hour, 2 hours, 6 hours and 24 hours. Samples are not discarded when a token drops off the ranking or when the user switches chains, and missing windows are backfilled from closed one-minute candles near the relevant timestamp, with the actual time and source stored. A rejected control group is sampled at roughly one fifth by hashed contract address, capped at 200 entries. The two groups report expiries, completions, missing counts and median change separately, never merged into one return figure. The README states plainly that missing price is not zero return, that the rejection sample only covers tokens actually deep-audited, and that 50 samples is a minimum observation threshold, not proof of strategy effectiveness.

Installing Meme Radar on Windows or macOS and running a first scan

The README points to two portable downloads on the Releases page, one for Windows x64 and one for macOS. Windows users unzip fully, then double-click MemeRadar-OpenSource.exe; the portable package bundles its runtime, so Node.js is not required. If SmartScreen intervenes, the README says to confirm the file came from the repository and then choose More info, then Run anyway. The black console window must stay open, because closing it stops the radar. If the browser does not open by itself, the local page is at http://127.0.0.1:3791/.

On macOS you unzip into a writable folder and double-click 安装并启动.command. First launch checks for Node.js, and if a compatible runtime is missing it downloads a project-specific build from nodejs.org, verifies SHA-256, installs pinned dependencies and opens the browser. If Gatekeeper blocks the script, right-click and choose Open.

If you already have Node.js 22.23 or newer, or 24.5 or newer, the repository also documents a command-line path:

npm run setup
npm run open

npm run setup installs dependencies from the lock file, and npm run open starts the local service in the background and opens http://127.0.0.1:3791/. For foreground debugging the README gives npm start, environment checking is npm run doctor, and tests are npm test. If the port is taken, RADAR_PORT selects another one. The README notes that the macOS auto-install path has been tested on real hardware, while Windows and Linux still need verification on clean machines.

Connecting GMGN is the step that makes scanning work. Click the first-use or create-API button, and the radar generates an Agent public key locally. Copy it, open the GMGN API creation page from the button, and paste it there. Enable read permission only and make sure trading permission is off. Then copy the resulting API key back into the radar and confirm. The key is written only to state/gmgn-api-key inside the project, and the Ed25519 authentication private key goes into a restricted state file with directory mode 0700 and file mode 0600. Every new API key requires the pairing step again; an old key's pairing cannot be reused.

Once scanning runs, tokens can be opened on the project website, GMGN or X for manual narrative and community checks. Favourites and notes are stored locally in state/preferences.json, capped at 50 favourites and 500 notes.

Where the local-only design breaks down

The security boundary is the selling point, and it is also the source of the tool's main operational weaknesses. The HTTP service listens only on the loopback address, which means there is no remote dashboard, no shared team view and no way to check the radar from a phone. Everything happens on the machine that runs it.

Availability is tied to that machine. The README states that shutdown or sleep means no scanning, and this version does not install a system startup item. A local process guard restarts the launcher after a crash, and state and settings files each keep one valid backup, but neither survives a powered-off laptop. If you want continuous coverage across a full day, this is the wrong shape of tool.

Rate limits are a real constraint rather than a footnote. The instant window shares the same single request queue and the same global 20-second interval as everything else, so it still consumes API quota. The README says outright that it cannot promise zero impact on audit speed. Each round of the pending-verification queue takes at most one priority slot and does not increase the total audit budget. That queue accepts at most 12 requests, expires after 10 minutes, and is cleared on restart.

There are also correctness limits the project documents against itself. The interface marks the second-source coverage for the current chain, and the README warns that "connected" does not mean every query succeeds. The instant window excludes known honeypots, wash trading, high risk, tokens created less than five minutes ago and records with insufficient base liquidity, but unknown risk is still labelled as unverified. The new-to-ranking flag means a token newly appeared relative to the last valid ranking, including re-entries, not that it was just launched. Price change between two snapshots uses the real interval and is not presented as one-minute candle performance. The statistics exclude tradability, slippage and fees, so they are not simulated trading returns. And if GMGN times out behind a VPN or proxy, the fix is to enable the proxy client's system proxy and fully restart the radar, not to retry inside the app.

Meme Radar Open Source compared with running GMGN and GoPlus in browser tabs

The obvious alternative is manual research: GMGN for discovery and labels, GoPlus for contract risk, DexScreener for liquidity and market cap, plus a spreadsheet for notes. That workflow has no install, no Node.js requirement and no local state to protect. Its weakness is that nothing is remembered consistently. Shadow tracking in particular is hard to do by hand, because you have to revisit the same contract address at 5 minutes, 15 minutes, 30 minutes, 1 hour, 2 hours, 6 hours and 24 hours, and backfill from closed candles when you miss a window.

Meme Radar's difference is that it automates the collection and the record-keeping while keeping the judgement human. It also keeps the data local, which browser tabs do not: the README states that browser storage never receives the API key or private key, that the page only gets state filtered through a field whitelist, and that upstream raw responses are not returned. A manual workflow can achieve similar hygiene only through discipline.

What the manual approach does better is breadth. Meme Radar reads GMGN's one-minute turnover ranking, up to 100 records, and its deep audit is capped at six tokens per round by default. A researcher working directly in GMGN can follow any pool, any wallet, any narrative, without a request budget. The radar also cannot use the GMGN CLI's other capabilities: the README states that the open-source edition does not call follow-wallet, swap or order endpoints, and that it will not read environment variables, an old global GMGN config or a project .env file. Only a key created and verified through the current open-source page takes effect. If your workflow depends on wallet-following, this edition does not replace it.

Licence, AGPL-3.0-only and what upgrading costs you

The source is licensed under GNU Affero General Public License v3.0, identified in package.json as AGPL-3.0-only. The README summarises the practical effect: you may use, study, modify and redistribute it, but if you modify it and offer it to others over a network, you must provide those users with the corresponding source under the licence. That is a stronger obligation than a permissive licence, and it applies to hosted derivatives, not just distributed binaries. This is a description of the licence text, not legal advice; read LICENSE and THIRD_PARTY_NOTICES.md before building a service on top of it.

package.json sets private to true, which the README explains is only there to prevent an accidental npm publish. The package name is meme-radar-open-source and the version is 0.1.6, matching the v0.1.6 release dated 2026-09-11. The repository's last push was on 2026-09-11, six days before this writing, so the project is current rather than abandoned, though a single release is thin evidence of a maintenance cadence.

Upgrade cost is low by design. There is one runtime dependency, gmgn-cli pinned at 1.5.7, and the radar uses only its read-only client. The Node.js engine range is ^22.23.0 or >=24.5.0. The macOS launcher validates a SHA-256 hash for the runtime it downloads, and npm run setup installs from the lock file. State lives in state/, logs in logs/, the auto-downloaded runtime in .runtime/, and the README says these plus node_modules/ must not be committed or shipped in a release archive. Third-party data endpoints remain subject to their own terms of service regardless of the AGPL grant, and docs/EDITION-BOUNDARY.md documents what separates this edition from the professional one.

Editorial conclusion

Adopt Meme Radar Open Source if you already review meme tokens by hand on Solana, BNB Chain, Base, Ethereum, Robinhood Chain, Arc or Stable, and you want the discovery, contract-risk and liquidity evidence collected in one local page instead of six browser tabs. Skip it if you expect a signal service, an automated entry point, or anything that trades: the README states the open-source edition contains no wallet private keys, no signing, no swap and no order module, and its three output tiers are reject, pending review and viewable, where viewable only means the token enters a manual research list. Before trusting a scan, confirm the GMGN key was created with read permission and trading disabled, check that state/ and logs/ are outside any repository or release archive, and read docs/EDITION-BOUNDARY.md so you know which capabilities the open-source edition deliberately leaves out.

Frequently asked questions

Official sources

1. Issues
2. License: AGPL-3.0
3. nhovongoc0-max/meme-radar on GitHub
4. README
5. Releases