How to Check If a Solana Token Is Safe with Claude and GMGN Skills

To check a Solana token with Claude, install GMGN Skills in Claude Code, configure a GMGN API key, enter the token’s mint address, and describe the checks you want in natural language. GMGN Skills queries the relevant token data, and Claude organizes the results into a readable risk summary that highlights current concerns and any data it couldn’t confirm. For example, you could ask: “Check the token security and pool status for [mint address] on SOL — don’t place a trade.”

For a more complete analysis, ask Claude to also cover holder concentration, creator holdings, and activity from insider, sniper, and bundler wallets. Even with these additional checks, no tool can prove that a token is “safe.” The result is a snapshot of the risks observable at the time of the query, not a guarantee about the project or its future price.

This guide uses Claude Code as the working example. The same workflow can also be used with other AI agents that support GMGN Skills.

Why check a new Solana token before buying

Important on-chain risk signals — unrevoked Mint or Freeze Authorities, low liquidity, concentrated holdings, and suspicious wallet and trading activity — can often be screened before a trade.

Solidus Labs analyzed Pump.fun tokens created between January 2024 and March 2025: of the more than 7 million with at least five trades, only about 97,000 maintained liquidity above $1,000. This doesn’t establish that any single token is fraudulent, but it shows how often new tokens end up too illiquid to trade or exit — which is why a current on-chain risk check matters before you buy (Solidus Labs, May 2025).

Check a Solana token for risk with GMGN Skills in Claude Code: 5 steps

Once GMGN Skills is installed and your GMGN API key is set, you can run the entire check in Claude Code with natural-language instructions. This token-checking workflow uses read-only queries and does not submit transactions.

Why does Claude need GMGN Skills? By default, Claude Code doesn’t have built-in access to GMGN’s current token data. GMGN Skills lets Claude Code use gmgn-cli to query GMGN’s API for current token security, pool, holder, and trader data; Claude then organizes the results into a readable explanation (source: GMGN Token Skill documentation).

Step 1 — Install GMGN Skills in Claude Code. Run:

npx skills add GMGNAI/gmgn-skills

This is the install command from GMGN’s official Quick Start for AI agents. If a later command reports that gmgn-cli isn’t found, install or update it with npm install -g gmgn-cli and try again.

Step 2 — Configure your GMGN API key. Generate an Ed25519 key pair, upload the public key at https://gmgn.ai/ai to create your API key, and save GMGN_API_KEY to ~/.config/gmgn/.env as described in the docs. The read-only token queries in this guide need only GMGN_API_KEY, not GMGN_PRIVATE_KEY. GMGN_PRIVATE_KEY is the private half of the authentication key pair that gmgn-cli uses to sign swap and order requests locally; it is not a blockchain wallet private key and doesn’t directly control on-chain assets, but authenticated trading requests can submit transactions that move real funds, and confirmed on-chain transactions can’t be reversed. The public key is safe to share, but never expose GMGN_API_KEY or GMGN_PRIVATE_KEY in chats, logs, screenshots, or version control. Full setup: GMGN docs.

Step 3 — Give Claude the full mint address. The mint address is a Solana token’s unique on-chain identifier. Don’t rely on a name or ticker — different tokens can share the same name and symbol, and only the mint address tells Claude exactly which token to analyze.

Step 4 — Describe the check in natural language. For a quick look, ask Claude to check the token security and pool status. For a broader assessment that also covers holders and wallet behavior, use the recommended prompt below (replace [mint address] with the token’s mint address):

Analyze this Solana token with GMGN Skills: [mint address]. Check
token authorities, liquidity, holder concentration, creator holdings,
and activity from `rat_trader`, `sniper`, and `bundler` wallets.
Classify each signal as lower concern, higher concern, or unavailable,
explain the evidence, summarize the main risks, and do not place a trade.

This is a recommended prompt, not a fixed format. A good prompt names the checks you want, asks Claude to mark missing data as unavailable rather than assuming it’s safe, and makes clear the task is analysis only. A quick security-and-pool question still returns a structured security summary, but it won’t necessarily cover holders and traders.

Step 5 — Read the summary. With the prompt above, Claude organizes GMGN’s token security, pool, holder, and trader results into a structured risk summary and provides an overall assessment.

What Solana token risks does GMGN Skills check?

GMGN Token Skill provides token security, pool, holder, and trader data. The main Solana-specific signals fall into four areas: token control, liquidity and launch stage, holder concentration, and wallet and trading behavior.

Token control: have the Mint and Freeze Authorities been revoked?

GMGN reports these authority statuses in the renounced_mint and renounced_freeze_account fields. If the Mint Authority hasn’t been revoked, its holder can still mint new tokens; if the Freeze Authority is still active, its holder can still freeze token accounts.

Mint Authority controls whether an account can mint additional tokens. For a meme coin that claims a fixed supply, an active Mint Authority means its holder — which could be the creator, a multisig, or a project account — can still mint additional tokens and dilute existing holders.

Freeze Authority controls whether an account can freeze a specific token account. A frozen account can’t move its tokens, so holders may be unable to sell. According to Solana’s official Set Authority documentation, setting the Mint or Freeze Authority to None permanently revokes that authority, and the change can’t be reversed.

For a typical meme coin, a Mint or Freeze Authority that hasn’t been revoked is usually a higher-concern signal — but it doesn’t prove fraud or a rug pull by itself. Stablecoins, RWAs (real-world assets), and similar tokens may keep these authorities for supply management, compliance, or asset protection.

Liquidity and launch stage: is the token still on the bonding curve?

If the token launched on a bonding-curve launchpad such as Pump.fun, first check whether it’s still on the curve or has graduated to an open DEX pool; from there, GMGN looks at liquidity depth, pool creation time, and the LP burn status.

A token still on the bonding curve hasn’t graduated to an open market or migrated pool, so post-migration LP burn checks may not apply yet. Once it migrates, check whether the pool has sufficient liquidity and review the LP burn status where applicable.

LP burn means the LP tokens — which represent a claim on the pool’s liquidity — have been destroyed, removing the holder’s claim on that portion of liquidity. But it only applies to the burned LP tokens, and it doesn’t stop the creator from selling their own tokens. Low liquidity isn’t proof of a scam on its own, but it can make a position harder to exit, increase slippage, and amplify the price impact of large sell orders.

Holder concentration: is supply concentrated in a few wallets?

Here the key figures are the top-10 holder share, dev-team holdings, creator balance, and suspected-insider holdings. If a few wallets hold a large share of supply, they can move the market sharply by selling at once. When reading the list, exclude identified pool and exchange addresses. A creator balance of zero isn’t automatically reassuring either — it can mean the creator kept nothing, or that the tokens were already sold or moved to other wallets.

Trading and wallet behavior: are there unusual patterns?

This is where trader data complements the holder snapshot. Look at the rug_ratio risk score, whether wash trading is flagged, the volume share tied to rat_trader and bundler wallets, and the sniper_count (wallets that bought at launch). In GMGN’s data these are distinct wallet tags — rat_trader (insider or sneak trading), sniper (bought at launch), and bundler (bot-bundled buying) — and they’re analytical signals, not verdicts, so weigh them against how much supply the wallets control, their funding sources, and their selling activity. Smart Money and KOL tags identify historically high-performing or well-known wallets; they don’t indicate that the token itself is safe.

Under GMGN Token Skill’s current rules, a rug_ratio above 0.3 falls in the high-risk range — but it is one input in a broader assessment, not a standalone verdict (source: GMGN Skills SKILL.md).

How to read Claude’s Solana token risk summary

Work through the summary signal by signal: for each check, look at the value, the evidence behind it, and how the recommended prompt classifies it — lower concern, higher concern, or unavailable. These three are reporting labels requested from Claude, not GMGN API fields or GMGN-issued safety ratings; “unavailable” means the data couldn’t be retrieved or confirmed and shouldn’t be read as low risk. The fictional example below (it doesn’t correspond to any real token) shows how the labels apply to fields GMGN Skills can return.

In this example, the Mint Authority is renounced, but the unrevoked Freeze Authority, low liquidity, concentrated holdings, and high rug_ratio create several higher-concern signals. Assess the pattern together rather than relying on any single favorable result. The reference points above (roughly $10k for low liquidity and 0.3 for rug_ratio) follow the GMGN Token Skill documentation.

What can a Solana token check still miss?

A GMGN Skills check is a snapshot of the data available when the query runs, not continuous monitoring or a safety certification. Liquidity, creator holdings, and wallet behavior can all change afterward, and an LP burn doesn’t stop the team from selling tokens held in other wallets. A soft rug is especially hard to catch: CertiK notes that this type of behavior does not require complex code and can be difficult to detect early, so a snapshot check can’t reliably predict whether a team will exit later (CertiK).

The check also can’t verify every off-chain factor, including team identity, claimed partnerships, paid promotion, compromised social accounts, or bot-driven community activity. Missing data should be marked unavailable rather than assumed safe, and wallet labels and the rug pull risk score are analytical signals — verify anything important through additional on-chain or independent sources.

Disclaimer: This article is for informational purposes only and is not investment, trading, or financial advice. No automated check can predict a token’s future price performance. Product capabilities, supported chains, field names, and risk references are current as of July 2026 and may change as GMGN Skills is updated. GMGN does not promise, guarantee, or predict any profit or return. Any user gains or case studies mentioned are individual, exceptional, and not representative of typical outcomes or of future performance.