Model Context Protocol

Shared context for teams building with Claude.

Every Claude session starts from scratch. In teams this compounds — each person re-explains the same project, makes decisions nobody else knows about, wastes tokens on warmup.

agntx fixes this. Context lives on a server. Your team's decisions, architecture, and current state are injected into every session automatically.

$npx @agntxapp/agntx init

How it works

01
Install
npx @agntxapp/agntx init detects your stack, pre-fills CLAUDE.md with known gotchas, and wires up the MCP server. Committed to git — every teammate gets it automatically.
02
Pull
Open Claude Code, type /status. The MCP server fetches your team's context from the server and injects it. Claude confirms understanding before touching anything.
03
Push
At session end, type /save. Claude structures what happened — decisions made, files touched, what's next — and pushes to the team. No writing required.

The loop

Developer A pulls context → does work → saves context
Developer B pulls context → sees Developer A's decisions → continues without re-explaining anything

4 commands

/status
Start of every session
Pulls context, confirms understanding. Ends with "Ready. What do you want to work on?"
/save
End of session
Claude extracts decisions and pushes to the team. You don't write anything.
/resolve
After a conflict
Two devs saved at the same time. Pick A, B, or both — logged automatically.
/diff
Before starting work
Shows what changed in the context since your last pull.

Stack templates

agntx detects your stack and pre-fills your CLAUDE.md with community-maintained gotchas and rules — the things that actually bit people in production.

Expo + Supabase
mobilerealtimeauth
Next.js + Supabase
app routerssrrls
Next.js + Prisma
app routerormmigrations

No template for your stack? The wizard interviews you instead. Contribute yours →

Why not just update CLAUDE.md

Every /save captures not just what changed, but why— and what was rejected. This is append-only. Nothing is ever overwritten. Six months later, when someone asks “why don't we use X here?” — the answer is already in the log.

// decisions_log entry
{
  "decision": "Use Supabase AI for embeddings",
  "reason": "Zero external dependencies, built into Edge Functions",
  "rejected_alternatives": [
    "OpenAI ada-002 — adds OpenAI dependency"
  ]
}

Pricing

Free during beta.

No credit card. No limits. We'll announce paid plans before charging anything.

Ready to stop re-explaining your project?

One command. Ships via git. No per-developer setup.

$npx @agntxapp/agntx init