GhostScanThe field guide
Get closer to the evidence.
GhostScan combines the preserved character instruments with a source-backed research workspace. Start with a contract, inspect the source coverage, then compare independent model perspectives.
01 / Find the market
Open Markets and choose Solana, BNB Chain or Robinhood Chain. Search by name or full contract, select a source preset, and inspect liquidity and source timestamps. Missing prices and stale history remain visible. Open an asset for seven OHLCV periods, trades, permissions and four-model analysis.
02 / Read the holder structure
Holder scan reads reported owners, mint permissions, risk sources and a bounded transaction sample. Full scans include funding and launch-window evidence by default. Expand the eight checks, search and sort wallets, inspect funding clusters, and request up to 100 transactions for a selected wallet. Read holder entries, whale balances, first buyers, developer launches and same-symbol assets. Export CSV, evidence JSON or a locally generated PNG / SVG report.
Reported shares use total supply, not verified tradable supply. Relationship edges are clues, not verified controller identities. Stablecoin entry estimates cover directly observed swaps; complete cost basis, historical entry market caps, wash-trade attribution and complete profitability remain unavailable where source data is missing.
Scanner data connections
Add SOLANA_TRACKER_API_KEY to this project’s server .env.local, restart the service, then run a fresh full scan. Use a Solana Tracker REST Data API key for full holder pages, strict token cost basis, wallet PnL, first buyers, developer history and bundlers. Basic artwork and mainnet evidence remain available from the existing market and Helius connections.
Solana Tracker Data API ↗ · Helius token metadata ↗
AutoScan is controlled by the operator. Enable continuous scans or set a daily limit, with an interval of at least one minute. It reads fresh indexed Solana new and trending pools, scans one coin at a time, and prioritizes unscanned coins. Repeat scans wait six hours. Status, current progress and the next candidate time appear above the list. It continues while the local service is running, even with the page closed; stopping the service stops scanning. Provider quota and rate limits still apply. It uses data credits without calling the four models or upgrading your plan.
Your Telegram scan bot
Set TELEGRAM_BOT_TOKEN, TELEGRAM_BOT_USERNAME, a random TELEGRAM_WEBHOOK_SECRET and an HTTPS QUORUM_PUBLIC_ORIGIN. Check with npm run setup:scan-bot -- --check; register your webhook with npm run setup:scan-bot when ready. A loopback-only local server cannot receive Telegram webhooks.
03 / Compare the council
GPT, Claude, DeepSeek and Grok read a shared snapshot. Choose Run analysis to call the configured providers. Answers, failures and conflicting judgments remain independent. The raw evidence, source times and model names are visible. Research history supports same-pool comparisons and PNG exports; private follow-up questions stay out of the public asset timeline.
04 / Observe new launches
Launch radar separates raw arrivals, indexed market data, reviewed snapshots and strict screened entries. Pool creation does not establish token creation. Admission requires fresh market data, complete covered risk fields and four valid model judgments. Robinhood metadata-only coverage cannot pass these gates.
05 / Follow changes and use the tools
Watchtower checks up to five public asset tasks while the page is visible. The workspace Tools screen provides persistent server rules, RSS/X, KOL bundles, asset dossiers, research, read-only agents, simulation plans, tracked-wallet proposals, historical backtests, keyboard and voice preferences. External notifications require your configured channel.
Paper accounts begin with zero funds. Add a simulation balance deliberately. Live quotes and transaction construction require your own wallet, allowance checks and final confirmation. Model tools cannot sign or broadcast; no private-key import is offered.
Connect your assistant
Start GhostScan, open Tools → Agents, and create a local read-only token. Replace YOUR_GHOSTSCAN_READONLY_TOKEN in your client configuration with that token. Keep this token private and separate from your model and Solana Tracker keys.
Use this site’s origin followed by /api/mcp/. The homepage copies the endpoint for the site you have open, including a local workspace. It provides HTTP commands and the exact configuration format for Claude, Codex, Cursor, Gemini and VS Code. Codex reads its Bearer token from GHOSTSCAN_MCP_TOKEN in the environment that starts Codex. Other examples show the Authorization header value you replace locally.
Tools start paused. Choose Resume read-only when ready; Revoke & stop invalidates previous tokens. External client setup is completed in your own assistant. The homepage only copies instructions and never registers clients or creates credentials automatically.
Start your local GhostScan
Use Node 24 or later and open a terminal in this project folder. Windows: .\scripts\start-tidal.cmd. Mac / Linux: sh scripts/start-tidal.sh. Both install missing dependencies, build the project and start port 4437. Stop an existing service on that port before starting another.
Once the browser workspace opens, use another terminal to run npm run cli -- doctor. Keys belong in the ignored .env.local; launchers keep those settings. --check verifies launcher prerequisites without starting another server.
Local CLI and administration
Run npm run cli -- doctor from this project. The server listens on 127.0.0.1:4437. The dedicated administration page manages project CA and X metadata. Local loopback access is separate from production authentication. Configure an exact production origin and operator password before a public deployment.
Saved instruments and provenance
The six character-tour scenes are dated interface samples. Original documentation describes the upstream CLI, while saved ScanIt reports preserve the captured feed, five report tabs, wallet profiles, entry clusters and sharing interactions. These archives are labelled historical and are not live backend results. License and source notices remain available with the code.
Questions
Why are some checks unknown?
The source may not cover the asset or provide complete ownership/history. An unavailable field never becomes zero or safe.
Does a successful model response verify its claims?
No. Models can misread prices, percentages, timestamps and account categories. Compare claims against the raw source snapshot.
Does the scanner use the original ScanIt risk engine?
No. That engine was not included in the supplied archive. GhostScan uses the configured mainnet sources and transparent checks documented in this guide. Historical ScanIt reports are kept separately.
Do observations run after I close the page?
Personal Watchtower tasks stop. Explicitly enabled server rules run while the local service is running. Automatic paid reviews start disabled in this new workspace.
Where is my data stored?
Public scan reports and workspace rules use this project’s own SQLite database. Credentials stay in the ignored server environment. The previous project’s database and deployment are unchanged.