teammate-sync 0.1.0

Cross-engineer Claude Code context sharing — type /share, your teammate queries your context.

teammate-sync

Cross-engineer Claude Code context sharing. Your teammate types /share in their Claude Code session. From your own terminal, your Claude can ask theirs anything and gets back a cited answer in seconds.

No pasted transcripts. No GitHub pushes. No shared bucket setup.

• Landing: https://omdivyatej.github.io/teammate-sync-site/
• Source: https://github.com/omdivyatej/teammate-sync

---

The problem

You and your teammate are both using Claude Code on the same project. You need to know what they decided about the database schema, or which file they touched, or why they picked one library over another.

Today, two bad options:

1. Ping them on Slack and wait. Forty-five minutes of context-switching.
2. Get them to send their session transcript. Their .jsonl is 50k lines. You paste it into your Claude and immediately blow past the 200k token context window.

teammate-sync is the third option: their Claude becomes a queryable context server. Your Claude calls it, gets back a cited synthesis, your context window stays clean.

Install

pip install teammate-sync
teammate-sync init
teammate-sync daemon

That's it. Full setup walkthrough: INSTALL.md.

init opens GitHub OAuth (your workspace = your GitHub org, no separate account), installs the /share slash commands into ~/.claude/commands/, wires session lifecycle hooks into ~/.claude/settings.json, and registers the MCP server. daemon runs in the foreground in a terminal you leave open.

How sharing works

/share                              in any Claude Code session

Daemon log immediately prints share-mode ACTIVATED -> uploading. That session's .jsonl plus your CLAUDE.md start syncing to your team's backend in real time as you work.

/unshare                            same session

Daemon wipes that session's content from the backend immediately.

/shared                             any session

Lists which sessions are currently shared.

How teammates query you

From any Claude Code session in your org, your teammate types:

Use mcp__teammate-sync__query_teammate_context with teammate=<your-github-handle> and question="..."

The MCP server pulls your active-session state + your shared session transcripts, hands them to a Claude Sonnet synthesis call, and returns a cited answer. Citations point to specific session IDs and file names so they can verify.

Privacy model

Default: nothing is shared. The daemon is running but idle.

• Sessions are opted in one at a time via /share. Unshared sessions never leave your machine, no matter how many you have running.
• Your GitHub OAuth token sits in ~/.teammate-sync/auth.json (mode 0600) and is never in the synced tree.
• Your org is the unit of trust. Only members of your GitHub org can query you. Discovery happens via the GitHub API, no manual invites.

What syncs where

Path	What it holds	Synced to cloud?
~/.teammate-sync/auth.json	Your GitHub OAuth token	NEVER (local only)
~/.teammate-sync/state/.shared-sessions.json	List of currently-shared session IDs	No (local gate)
~/.teammate-sync/state/.active-sessions.json	Which sessions are active right now (cwd, last activity)	Yes
~/.claude/projects/<encoded-cwd>/<sid>.jsonl	Claude Code session transcripts	Only /share'd ones

CLI

teammate-sync init	First-run setup. Re-runnable to refresh hooks / slash commands.
teammate-sync daemon	Run the sync daemon (foreground).
teammate-sync share / unshare / shared	Same as the slash commands, from any shell.
teammate-sync whoami	Show your identity + workspace.
teammate-sync teammates	List members of your workspace org.
teammate-sync logout	Delete ~/.teammate-sync/auth.json.

Requirements

• Python 3.11+
• Claude Code CLI
• ANTHROPIC_API_KEY exported in your shell (used by the MCP server's synthesis calls)

Status

Public beta. Backend runs on Fly.io (Singapore region, single-region by design — synthesis latency is dominated by Anthropic API not by storage). Per-session opt-in (/share / /unshare) is the privacy model; there is no "everything sync" mode and there won't be.

License

MIT — see LICENSE if present, or pyproject.toml.