Plan, review, execute

PlanDrop

Talks to Claude Code from your browser

A Chrome extension that lets you plan, review, and run Claude Code tasks on remote servers. No terminal required.

Add to Chrome Documentation GitHub
PlanDrop Extension
📋 PlanDrop
Run DESeq2 differential expression on treated vs control...
terminal
$ plandrop-watch
[PlanDrop] Watching for plans...
[PlanDrop] ✓ Using Claude login (Max)
[PlanDrop] Task received, running...

How It Works

Four steps from idea to execution

1

Start the watcher on your server

terminal
$ cd ~/rnaseq_project
$ plandrop-watch
[PlanDrop] Watching for plans...
[PlanDrop] ✓ Using Claude login (Max subscription)
2

Open PlanDrop side panel, type your task

"Create a DESeq2 differential expression analysis.
Compare treated vs control using the count matrix in data/.
Apply proper normalization, shrinkage, and FDR correction."
3

Review Claude's plan in the activity feed

Claude: I'll create the analysis with these steps:
1. Load count matrix and sample metadata
2. Create DESeqDataSet with ~condition design
3. Run DESeq2 with lfcShrink (apeglm)
4. Filter FDR < 0.05, |log2FC| > 1
5. Generate volcano plot and MA plot
Write: scripts/deseq2_analysis.R
Write: scripts/run_deseq2.sh
4

Click Execute → watch it run → see results

✅ Complete · Free (Max) · 23.4s
Files created: deseq2_analysis.R, volcano_plot.pdf, ma_plot.pdf
1,247 significant genes (FDR < 0.05)

Two Modes for Different Needs

Choose the right mode for your task

💬

Claude Code

Interactive

Full interactive sessions with plan-review-execute workflow. Best for complex tasks that need iteration.

  • ✓ Session continuity (--resume)
  • ✓ Multi-file operations
  • ✓ Full activity feed
  • ✓ Execute commands
📝

Quick Drop

Fast

Drop a file and get instant analysis. Best for quick code reviews and one-off questions.

  • ✓ Drag-and-drop file upload
  • ✓ Instant analysis
  • ✓ Plan-only (read-only)
  • — No execution

Everything you need

Plan your tasks in the browser, execute on your server

🧠

Plan → Review → Execute

Claude Code must produce a read-only plan first. You review before authorizing. Human-in-the-loop by architecture, not by prompting.

🛡️

Permission Profiles

Domain-specific: Bioinformatics, ML/Deep Learning, Edit Files Only, Full Access, or custom.

📝

Prompts as Files

Every prompt saved as a .md file on the server. Git-trackable, re-readable, re-sendable.

🔄

Session Continuity

Claude Code remembers context across tasks via --resume. Build on previous work without re-explaining.

📊

Activity Feed

See Claude's reasoning, file writes, bash commands, blocked actions, and costs in real time.

🖥️

Multi-Project Dashboard

Manage multiple servers and projects from one side panel. Per-tab project binding.

🔒

Zero Infrastructure

No WebSocket server, no database. Just SSH + files. All communication stays on your local network.

💰

Uses Your Max Subscription

No API costs. Works with Claude Max login. Free execution.

🔔

Browser Notifications

Get notified when long tasks complete while you're in another tab.

📋

Full History Export

Export complete task history as markdown. Every request, response, file change, and command logged.

Your workflow, transformed

Stop copy-pasting between ChatGPT and your terminal

😩 Without PlanDrop
1 Open ChatGPT, describe your analysis
2 Copy the code response
3 SSH to server
4 Paste into vim/nano
5 Fix formatting issues
6 Run the script
7 Copy error back to ChatGPT
8 Repeat 5 times
9 No record of what you asked
😎 With PlanDrop
1 Type task in side panel
2 Review Claude's plan
3 Click Execute
4 Done. Prompt saved as .md, response logged, git-trackable.
⚡ 4 steps instead of 9

Permission Profiles

Control exactly what Claude Code can do on your server

📋

Plan Only

Read files, analyze code, suggest changes. No writes, no commands.

📝

Edit Files Only

Read and write files. All shell commands blocked.

Default
⚡

Standard

Full access except dangerous commands (sudo, rm -rf /, shutdown...).

Caution
🔓

Full Access

No restrictions. Requires confirmation. Use in sandboxed environments only.

⚙️

Custom

Define your own blocked commands using deny-list patterns.

Learn more about permission profiles

Designed with Safety in Mind

Human reviews every plan before execution

✅

You Review Every Plan

Claude Code must produce a read-only plan. You approve before any execution.

🛡️

Permission Profiles

Restrict what Claude Code can do. Domain-specific or custom.

🔒

Zero Infrastructure

No WebSocket server, no database. Just SSH + files. Data stays on your network.

📂

Fully Open Source

Audit every line of code. No black boxes.

Learn more about trust & safety

How It All Connects

Simple architecture, no infrastructure to manage

🌐

Browser (Chrome Side Panel)

You type tasks in the PlanDrop extension

Native Messaging
💻

Local Machine (plandrop_host.py)

Native host handles SSH connections

SSH (ControlMaster)
🖥️

Remote Server (.plandrop/)

Plans written to files, watched by plandrop-watch

watch.sh
🤖

Claude Code (plan → execute)

Plans, then executes after approval

JSONL response
🌐

Back to Browser (activity feed)

Results streamed back to you in real time

Get started in 2 minutes

Simple setup on server and browser

1

Install the watcher on your server

curl -sL https://raw.githubusercontent.com/genecell/PlanDrop/master/server/setup.sh | bash
2

Initialize your project directory

cd your-project && plandrop-watch --init
3

Start the watcher

plandrop-watch

Tip: Use tmux to keep it running in the background.

Ready to get started?

Check out the documentation for detailed guides and reference.

Read the Docs View on GitHub