context-share

context-share

Share conversation context between Claude Code agents across sessions and machines.

One agent runs /context-share:save my-key, another runs /context-share:load my-key — on any machine — and picks up right where the first left off.

How It Works

Two components:

1. Server — A small Go HTTP server that stores and retrieves context bundles, backed by SQLite.
2. Plugin — Three Claude Code slash commands (/context-share:save, /context-share:load, and /context-share:delete) that compile, transfer, and manage context.

The server is a self-contained binary with no external dependencies. SQLite is embedded — no database server to install or configure. The database file and table are created automatically on first run.

Server Setup

Requirements

• Go 1.22+ (to build from source), or just grab the binary

Build

cd server
go build -o context-share .

Run

export CONTEXT_SHARE_API_KEY="your-secret-api-key"  # required
export PORT="8787"                                    # optional, default: 8787
export DB_PATH="context-share.db"                     # optional, default: context-share.db

./context-share

The server will create context-share.db in the current directory on first run. No migrations or initialization needed.

Plugin Setup

Install the plugin

# 1. Add the marketplace (from GitHub)
claude plugin marketplace add sam-david/claude-code-context-share-plugin

# Or from a local path if you've cloned the repo
claude plugin marketplace add /path/to/claude-code-context-share-plugin

# 2. Install the plugin
claude plugin install context-share

Configure environment variables

Add to your shell profile (~/.zshrc, ~/.bashrc, etc.):

export CONTEXT_SHARE_URL="http://your-server:8787"    # or https:// if behind a reverse proxy
export CONTEXT_SHARE_API_KEY="your-secret-api-key"     # same key the server uses

Both variables are required on every machine that will use the plugin.

Usage

Save context

In any Claude Code session:

/context-share:save my-feature

The agent compiles a structured context bundle from the conversation — summary, decisions made, files touched, current state, next steps, key code snippets, and warnings — then uploads it to the server under the key my-feature.

Save only specific context instead of everything:

/context-share:save my-feature the database schema and why we chose SQLite

Optional TTL (auto-expire after N hours):

/context-share:save my-feature --ttl 72

Load context

In a new Claude Code session (same machine or different):

/context-share:load my-feature

The agent fetches the context, presents it, reads relevant local files if they exist, and offers to continue where the previous session left off.

Delete context

/context-share:delete my-feature

Removes the context from the server.

Deployment Options

Cross-compile for deployment

Build on your Mac, deploy to a Linux server:

GOOS=linux GOARCH=amd64 go build -o context-share .
scp context-share your-server:/opt/context-share/

Run with systemd (Linux)

Create /etc/systemd/system/context-share.service:

[Unit]
Description=context-share server
After=network.target

[Service]
ExecStart=/opt/context-share/context-share
Environment=CONTEXT_SHARE_API_KEY=your-secret-api-key
Environment=DB_PATH=/opt/context-share/data/context-share.db
WorkingDirectory=/opt/context-share
Restart=always

[Install]
WantedBy=multi-user.target

Then:

sudo systemctl enable --now context-share

Docker

cd server
docker build -t context-share .
docker run -p 8787:8787 -e CONTEXT_SHARE_API_KEY=your-secret -v context-data:/data -e DB_PATH=/data/context-share.db context-share

API Reference

All endpoints require Authorization: Bearer <api-key> header.

Method	Path	Description
PUT	/context/{key}	Save context. Body: {"context": <any JSON>, "ttl_hours": <int or null>}
GET	/context/{key}	Load context. Returns {"key", "context", "created_at", "expires_at?"}
DELETE	/context/{key}	Delete context.
GET	/health	Health check. Returns {"status":"ok"}

Architecture Notes

• SQLite with WAL mode — handles concurrent reads/writes safely, no external database needed
• Constant-time auth comparison — prevents timing attacks on API key
• Expired contexts are cleaned up lazily on next read (no background job needed)
• Context is stored as arbitrary JSON — the server is schema-agnostic, all structure is defined by the plugin commands
• 14MB binary, ~5MB RAM — runs comfortably on the smallest VPS or even a Raspberry Pi