Using trenchscope.

Install.

One line. You need a terminal; the installer handles Node.

macOS and Linux

curl -fsSL https://raw.githubusercontent.com/trenchscopeapp/trenchscope/main/install.sh | bash

Windows PowerShell

irm https://raw.githubusercontent.com/trenchscopeapp/trenchscope/main/install.ps1 | iex

With Node 22.18 or newer

npm i -g trenchscope

The installers are user-level: no sudo and no edits to your shell profile. On macOS and Linux a checksum-verified Node goes into ~/.dexterm/node if yours is missing or too old. On Windows the installer offers Node LTS through winget. Read them first if you like: install.sh, install.ps1.

Try it before installing

Sample data, nothing written to your machine:

npx trenchscope --fixtures

Let your agent install it

Paste this into Claude Code, Codex or any coding agent:

Install trenchscope (https://github.com/trenchscopeapp/trenchscope): run `npm i -g trenchscope` (if Node 22.18+ is missing, use the one-line installer from its README instead), then run `trenchscope doctor` and fix anything it reports, then connect its read-only MCP server to yourself with `trenchscope mcp install claude-code` (or `codex`). Tell me to run `trenchscope` in my terminal when you are done.

First run.

trenchscope doctortrenchscope

doctor checks Node, your terminal's colour support, the optional claude or codex CLI, the network and every data source, then prints the fix for anything missing. trenchscope opens the screener on live data. Every run starts on paper.

The first launch plays the full boot sequence; later ones take a second. Any key skips it, and --no-boot leaves it out.

The app needs a real terminal. For a single frame printed to stdout, use trenchscope --screen screener.

Updates

trenchscope update checks npm and GitHub for a newer version and prints the upgrade command. It also checks once a day on start; trenchscope update --off turns that off.

Screens.

Filter, sort and presets

Press / in the screener and type plain thresholds:

vol24 > 250K liq > 50K age < 30d launchpad:pump

Sort s

vol liq mcap price age m5 h1 h6 h24 holders org top10 rug

Presets []

trench trending runners organic graduating smart traded new boosted all

The same work from the command line:

trenchscope --preset runners --filter "liq > 50K" --sort h1:desc

Open any coin

Press g on any list or the token screen and paste a contract address, or a DexScreener, pump.fun, Solscan or Birdeye link. Pasting one into the screener's / line or the analyst chat works too. A DexScreener link usually names the pool, so trenchscope finds its coin first.

trenchscope --screen token --mint <CA or link>

Keys.

Press ? on any screen for every key that works there. trenchscope keymap prints them all, and you can rebind them in ~/.dexterm/config.json.

The analyst.

Press a and your own logged-in Claude Code or Codex reads the same tool results the screen shows, through trenchscope's read-only MCP server, and answers in plain language. Usage is billed to your account. Everything else works without it.

• Every answer is labelled model output, with the age of every number it used.
• Token names, descriptions and fetched pages reach the model wrapped as untrusted text.
• It has no build, sign or send tool. The most it can do is navigate and open a paper ticket.

It uses claude if installed, else codex. Pick one with --analyst-engine claude or codex, or /engine in the chat.

Connect your agent.

trenchscope ships a read-only MCP server: your agent gets the same token dossiers, holder, rug and LARP checks, candles and trades the screener shows. It has no tool that can build, sign or send a transaction.

Claude Code

claude mcp add trenchscope -- npx -y trenchscope mcp

Codex

codex mcp add trenchscope -- npx -y trenchscope mcp

Claude Desktop, Cursor and others

trenchscope mcp install claude-desktop

Swap claude-desktop for cursor, codex, claude-code or generic. It asks before writing, backs up the old config and is safe to run twice. As a Claude Code plugin: /plugin marketplace add trenchscopeapp/trenchscope, then /plugin install trenchscope@trenchscope.

Claude.ai

Open Settings, Connectors, Add custom connector, name it Trenchscope and use this URL:

https://trenchscope-mcp-production.up.railway.app/mcp

Then ask Show trending Solana tokens, or Open a dossier for this mint with an address. No login or API key. The hosted server has 17 public analysis tools; screen navigation, paper trading and your history stay local. It shares a request budget, so a busy server answers HTTP 429 with Retry-After.

trenchscope mcp doctor checks the connection.

Paper and live trading.

Experimental. Paper and live trading are the roughest parts and have known bugs. Do not rely on them yet. Nothing here is financial advice.

Every run starts on a simulated wallet priced from live Jupiter quotes. On a token, B or S opens a paper ticket and P shows your positions.

Going live

Link a Phantom wallet first. trenchscope stores only its public key; you sign a message to prove it is yours, with no transaction.

trenchscope wallet link

Then L on a ticket flips it to LIVE. A separate signer process quotes, builds, simulates and checks the caps, and Phantom asks you to approve in your browser. It is devnet unless you start with --mainnet, which shows a red banner.

Raising a cap above its default needs --mainnet. trenchscope wallet panic disables live signing at once; trenchscope wallet resume turns it back on and needs a keypress. --no-live never starts the signer.

trenchscope never sees your private key or seed phrase, never signs without your keypress and your wallet's approval, and never gives the analyst or an agent a way to trade.

Data and keys.

Market data comes from public, keyless APIs, read from your machine. When a source fails, the cell says so instead of guessing.

unavailable the source did not answerneeds key the field wants a key you have not set

Add a Helius key

A free Helius key unlocks the ranked holder list, sniper and bundle scans and a faster on-chain lane. Keep it in a file, never in a command you share:

printf %s "<your-key>" > ~/.dexterm/helius.key && chmod 600 ~/.dexterm/helius.key

Any read-only RPC works too: set DEXTERM_RPC_URL. --max-credits caps Helius credits for a run (default 5000) and --max-requests caps every live request.

What it sends and stores.

No account, no telemetry, no crash reports. Local servers bind 127.0.0.1 with per-use tickets.

Over the network

• Requests to the public data APIs: the addresses you look at, and your IP.
• WebSocket subscriptions to the new-pairs feed and the RPC.
• A token's website, read-only, only when you press a link key.
• Your analyst prompts, through your own claude or codex login.
• When you trade live: the signer's quote, price and RPC requests and the transaction your wallet signed.
• A once-a-day version check: two anonymous requests to npm and GitHub. trenchscope update --off stops it.

trenchscope doctor lists every host and what it is for.

On your machine

A SQLite database with the cache, watchlist and paper wallet, a small state file in ~/.dexterm, the optional RPC key, and, if you use them, plan and copy journals, the signer's ledgers and your linked wallet's public key. All of it is yours to delete.

Make it yours.

When something breaks.

Setup failed or a source is down

Run trenchscope doctor. It prints the exact fix for anything missing.

Your agent cannot see trenchscope

Run trenchscope mcp doctor.

A field says needs key

It wants a Helius key. See Data and keys.

Your antivirus flagged a release

Every release ships checksums, a signed build attestation and third-party scan results. Verify your download.

You found a bug

Open an issue with your trenchscope doctor output; it holds no secrets. Security problems go to the security policy, not an issue.