Table of Contents
- 1. Why Asking an AI "Is This a Rug?" Usually Fails
- 2. What MCP Is, in Plain English
- 3. Setup: Claude, Claude Code, Cursor and VS Code
- 4. What the Rug Checker Looks At
- 5. How to Read the Risk Score
- 6. Prompts That Work Well
- 7. For Token Creators: How to Pass the Check
- 8. Limits You Should Know
- 9. FAQ
Why Asking an AI "Is This a Rug?" Usually Fails
Large language models are trained on text that was frozen months before you use them. A memecoin that launched this morning simply doesn't exist for the model. When you ask about it anyway, you tend to get one of three answers:
- A generic checklist ("check the liquidity, check the team…") that doesn't tell you anything about this token.
- A refusal to comment on a specific token.
- A made-up answer: a liquidity figure, a holder count or a "the mint authority is revoked" that the model never actually checked.
The last one is the dangerous one. Rug pulls on Solana depend on a handful of on-chain facts: who can still mint tokens, who can freeze your account, whether the liquidity can be pulled, and how much of the supply sits in a few wallets. Those facts change minute by minute and have to be read from the chain. A model can't know them without a tool.
The fix: give the assistant a tool that reads the chain, and let it do what it's good at: explaining the result in plain language and answering your follow-up questions.
What MCP Is, in Plain English
The Model Context Protocol (MCP) is an open standard for plugging tools into AI assistants. You can think of it as a USB port for AI: a tool provider publishes an MCP server once, and any assistant that speaks MCP can use it. That includes Claude, Claude Code, Cursor, VS Code and a growing list of agents.
When you ask a question the tool can answer, the assistant calls it, gets structured data back, and builds its answer from that data instead of from memory. You see which tool was called and with which input, so you can check what the AI actually looked at.
The CreateMyCoin Solana Rug Checker MCP server exposes exactly one tool, check_solana_token. It takes a token mint address and returns the same report as our web rug checker. It is read-only: it can't connect a wallet, sign anything or trade.
Setup: Claude, Claude Code, Cursor and VS Code
The server is remote, so there is nothing to install and no API key. All you need is the URL:
https://createmycoin.app/mcpClaude (web and desktop)
Open Settings → Connectors → Add custom connector, name it Solana Rug Checker and paste the URL. Start a new chat and make sure the connector is switched on in the chat's tools menu.
Claude Code
Run this once in your terminal:
claude mcp add --transport http solana-rug-checker https://createmycoin.app/mcpCursor
Add this to ~/.cursor/mcp.json:
{ "mcpServers": { "solana-rug-checker": { "url": "https://createmycoin.app/mcp" } } }VS Code
Add this to .vscode/mcp.json in your project:
{ "servers": { "solana-rug-checker": { "type": "http", "url": "https://createmycoin.app/mcp" } } }Any other client that supports remote MCP servers over Streamable HTTP works the same way: point it at the URL. For more details and copy buttons, see the MCP server page.
What the Rug Checker Looks At
Every check reads live on-chain and market data. Here is what is covered and what a red flag looks like for each:
| Check | What it means | Red flag |
|---|---|---|
| Mint authority | Whether someone can still create new tokens | Still active: the creator can inflate supply and dump |
| Freeze authority | Whether someone can freeze holders' token accounts | Still active: your tokens could be locked so you can't sell |
| Metadata | Whether name, symbol and image can still change | Mutable: the token can be silently rebranded |
| LP burned | Share of the pool's LP tokens that is burned | Under 50%: most of the liquidity can be withdrawn |
| Liquidity | USD depth of the main pool | Under $1,000, or no pool at all |
| Liquidity vs market cap | How much real money backs the valuation | Under 1%: price can collapse on modest selling |
| Top 10 holders | Supply share of the 10 largest wallets (pools and burned tokens excluded) | Over 50% |
| Creator holdings and activity | How much the deployer still holds, and whether they sold | Over 20% of supply in the creator's wallet |
| Volume, pool age, track record | Whether there is real trading and how long the token has survived | First pool under 24 hours old, almost no volume |
Pools on Raydium, PumpSwap, Orca and Meteora are recognised, as are pump.fun and other launchpad bonding curves. A token still on a bonding curve gets a pass on LP burn, because the liquidity sits in the launchpad contract and the creator can't pull it. For background on each signal, see our Solana rug checker guide and the list of token red flags.
How to Read the Risk Score
Every token starts at 100. Each failed check subtracts points, and the most serious findings cap the score no matter how good everything else looks:
| Score | Rating | What to do |
|---|---|---|
| 80–100 | Low risk | No serious on-chain problems found. Still do your own research on the team and community. |
| 50–79 | Caution | Read every failed check. Often a new pool, thin liquidity or concentrated holders. |
| 0–49 | High risk | At least one critical problem, such as an active mint or freeze authority or pullable liquidity. |
A few rules worth knowing:
- An active mint or freeze authority caps a new token at 40 (high risk). Established tokens with deep liquidity and months of trading get a softer penalty, but they never reach "low risk" while an authority is on.
- A pool under 24 hours old caps the score at 75. Most rugs happen on day one, so a brand-new token can't earn a low-risk rating, however clean it looks.
- Missing data costs no points but lowers the confidence value. A score of 85 with low confidence means several checks couldn't run. Ask your assistant which ones.
Unknown is not safe. If a check comes back as unknown (for example, the creator wallet couldn't be identified), treat it as an open question, not a pass. A well-behaved assistant will say so; if yours doesn't, ask it directly.
Prompts That Work Well
You don't need special syntax. Paste the mint address (the token's address, not a wallet or pool address) and ask in your own words. Some prompts that get useful answers:
- "Is EKpQGSJtjMFqKZ9KQanSqYXRcF8fBopzLHYxdM65zcjm a rug?"
- "Rug-check this token and explain every failed check like I'm new to crypto: <mint>"
- "Compare these three tokens on authorities, LP burn and holder concentration, then rank them by risk: <mint 1>, <mint 2>, <mint 3>"
- "What would this project have to change on-chain to reach a low-risk rating? <mint>"
Follow-ups are where the AI earns its keep. Ask why the creator's holdings matter, what "LP burned" means for you as a seller, or whether a freeze authority on a stablecoin is normal (it is: issuers like Circle keep it on purpose, and the checker recognises those issuers).
Where to find the mint address: on DexScreener, Birdeye or Solscan the token page shows it, usually with a copy button. Always take it from an official source. Scammers clone token names and logos, but they can't clone the mint address. Our guide on using DexScreener shows where to look.
For Token Creators: How to Pass the Check
If you're launching a token, assume buyers will run exactly this kind of check, increasingly through their AI assistant, before they buy. A token that comes back "high risk" because of a setting you forgot to change loses buyers who would otherwise have bought. The fixes are almost all done before or right at launch:
1. Revoke mint authority
Removes the single biggest penalty. See how to revoke mint authority.
2. Revoke freeze authority
The second hard cap. See how to revoke freeze authority.
3. Make metadata immutable
A small penalty, but a visible trust signal. CreateMyCoin does this for free at creation.
4. Burn or lock your LP tokens
95% or more burned counts as a full pass. Read what locked liquidity is and how much liquidity to add.
5. Don't sit on a huge share of supply
Keep the creator wallet at 5% or less and spread team or treasury allocations transparently, ideally behind a Squads multisig or a vesting contract.
On CreateMyCoin, steps 1 to 3 are checkboxes in the creation form, so a token can launch fully locked in one transaction (0.3 SOL plus 0.05 SOL per authority revoked). Then run your own token through the checker before announcing it, and post the result. The full pre-launch list is in our Solana token security checklist.
Limits You Should Know
- It's not an audit. The checker reads on-chain state. It can't tell you whether the team will keep building, whether the socials are real, or whether a coordinated group is about to dump.
- Solana SPL tokens only. It doesn't check EVM tokens.
- Data is cached for a few minutes. On a token that is moving fast, liquidity and holder numbers can lag slightly.
- Fair use is 30 requests per minute per IP. That's plenty for personal use. Comparing a few dozen tokens in one go may need a short wait.
- The AI can still misread the result. The report includes the raw checks, so if an explanation sounds off, ask the assistant to show the exact check and its detail.
Never share a seed phrase or private key with an AI assistant, or with any "rug checker", "wallet validator" or support bot. A rug check only ever needs the token's public mint address. Anything that asks for more is a wallet drainer.
The score is an automated estimate, not financial advice. Use it to filter out the obvious traps quickly, then spend your own research time on the tokens that pass.
FAQ
Can Claude check if a Solana token is a rug pull?
Not on its own: without a tool, an AI model has no live blockchain data and can only guess. Once you connect a rug checker MCP server such as https://createmycoin.app/mcp, Claude can read the token's authorities, liquidity pool, holders and creator wallet on-chain and explain the risk score in plain language.
Is the CreateMyCoin rug checker MCP server free?
Yes. It is free, needs no signup and no API key. Fair use is 30 requests per minute per IP address.
Which AI tools can use the rug checker MCP server?
Any MCP client that supports remote servers over Streamable HTTP. Tested setups include Claude on the web and desktop (custom connectors), Claude Code, Cursor and VS Code.
Does a high risk score mean a token is safe?
No. A score of 80 or more means none of the automated checks found a serious problem, not that the token is safe. It is not an audit and does not cover off-chain risks like a team abandoning the project. Treat it as a first filter, not financial advice.
Can the MCP server buy or sell tokens or access my wallet?
No. It exposes one read-only tool, check_solana_token, which takes a mint address and returns a report. It never connects a wallet, signs transactions or asks for keys. Never share a seed phrase or private key with any AI assistant.