Your AI coding tools forget.
Agent Stash remembers.

Shared memory across Claude Code, Cursor, and your team. One MCP server. No re-explaining.

Install How it works

Sessions die.

Long refactor, context fills, agent starts over. Memory persists across sessions so work isn't lost.

Tools are siloed.

Claude Code doesn't know what Cursor learned. Same project memory, both tools, no duplication.

Teams duplicate context.

Your teammate re-explains the architecture to their agent. Shared project memory ends that.

Get started in three steps.

Step 1

Get an API key.

Sign in with GitHub, or register headlessly in the terminal:

curl -X POST https://agentstash.ai/register/agent \ -H 'Content-Type: application/json' \ -d '{"agent_name": "my-project"}'

The response includes your api_key and a claim_url. Save both. Open the claim URL in a browser anytime to connect your GitHub account and enable upgrades — your key and all data carry over.

Step 2

Add to your MCP config.

Claude Code — add to ~/.claude/settings.json:

{ "mcpServers": { "agent-stash": { "command": "npx", "args": ["-y", "@agentstash/mcp"], "env": { "AGENT_STASH_API_KEY": "sk_your_key_here" } } } }

Cursor: same block in ~/.cursor/mcp.json. Memory is scoped to your git repo automatically.

Step 3

Use it.

Claude Code saves a decision:

remember("auth-approach", "Using session cookies, not JWT. See ADR-003.")

Cursor picks it up — new session, different tool:

recall("auth-approach") → "Using session cookies, not JWT. See ADR-003."

Works in Claude Code, Cursor, Codex, and any tool that supports MCP. Read the docs →

Pricing

Start free. Upgrade when you need more capacity.

Free

$0/mo

  • 5 memories
  • 3 persistent memories
  • 24h max TTL
  • 5 logs
  • 10 req/min
Get started

Pro

$30/mo

  • 25 memories
  • 10 persistent memories
  • 7 day max TTL
  • 50 logs
  • 60 req/min
Upgrade

Enterprise

$150/mo

  • 100 memories
  • 50 persistent memories
  • 30 day max TTL
  • 500 logs
  • 300 req/min
Upgrade