Your agent keeps forgetting.
Memory that works everywhere.

Persistent memory for AI agents. One account, every Claude surface.

$npx -y @clawtex/mcp-server

Works with Claude Code, Claude Desktop, and any MCP client

Built for the Anthropic Connectors Directory

Claude.ai·Desktop·Mobile·Claude Code

One account. Every surface. Memory follows the conversation.

The problem

You've been here before.

You re-explain the same context every session.

Your agent repeats mistakes you already corrected.

Last week's decisions vanish by Monday.

It's not your agent's fault. It has no persistent memory.

The fix

Three layers. One system of record.

State

What exists right now. Clients, projects, tools, repos. Structured and queryable. Loaded automatically at every session start, on every surface. Types are created dynamically based on what your agent stores.

Events

Everything that happened, logged as it happens. Decisions, milestones, blockers, tasks, deployments. Each event is timestamped, typed, and linked to an entity. Full history, fully queryable -- from any Claude session.

Lessons

Patterns extracted from your event history. Not vague suggestions -- structured guardrails with numbered steps and enforcement mechanisms. Injected at every session start. Your agent doesn't just remember. It improves.

Getting started

Connect in seconds.

1

Connect

Find Clawtex in your Claude settings. Click Connect. Authorize with your Clawtex account. Done. Works on Claude.ai, Desktop, and Mobile.

terminal
// Claude.ai / Desktop / Mobile:
// Settings > Connectors > Search "Clawtex" > Connect
//
// OAuth handles authentication automatically.
// No config files. No API keys to paste.
//
// Claude Code (CLI):
// claude mcp add clawtex --transport streamable-http \
//   --url https://mcp.clawtex.io/api/mcp
2

Or install locally

Prefer a local MCP server? Add the npm package to Claude Code or Claude Desktop. Same 15 tools, same memory.

terminal
// Claude Code / Desktop config:
{
  "mcpServers": {
    "clawtex": {
      "command": "npx",
      "args": ["@clawtex/mcp-server"],
      "env": {
        "CLAWTEX_API_KEY": "tkr_..."
      }
    }
  }
}
3

Build

Start a new session. Your agent loads state, logs events, and updates context automatically. Memory builds with every session.

terminal
// State and lessons auto-load at session start.
// Your agent also has tools for:
//
// clawtex_log_event - log decisions, milestones, tasks
// clawtex_update_state - track projects, clients, goals
// clawtex_get_events - query your full history
//
// No curl commands. Native tool calls.
4

Extract

Clawtex analyses your event history and extracts lessons from recurring patterns. Each lesson includes numbered correction steps and an enforcement mechanism. Approve or reject. Your agent improves every week.

terminal
// From any Claude session:
// "Extract lessons from this week's events"
//
// Clawtex proposes lessons like:
// [L001] When deploying code changes
//   CONTEXT: Skipped visual audit before submission
//   CORRECTION:
//     1. Run full visual audit checklist
//     2. Verify all screens on device
//     3. Check edge cases before marking complete
//   ENFORCEMENT: Checklist before submission

Architecture

Not a shared notepad. A system of record.

Most memory tools dump text into a shared context. Clawtex separates concerns.

Cold-start

Every session starts from zero. No memory of what happened yesterday, last week, or last month.

Solved by

Bootstrap. One API call at session start loads full structured context. State tells the agent what exists. Events tell it what happened.

Drift

The agent makes the same mistakes. Lessons from last month are forgotten. Quality degrades over long horizons.

Solved by

Lessons. Clawtex analyses event history, extracts patterns from recurring mistakes, and injects structured guardrails at every session start.

State tells your agent what exists. Events tell it what happened. Lessons tell it what to do differently. Each layer has a job. Together they compound.

The difference

Before and after.

Before Clawtex

Re-explain context every session

Agent repeats the same mistakes

Decisions lost after compaction

Context resets when you switch devices

No improvement over time

After Clawtex

Agent starts every session with full context

Mistakes caught and corrected automatically

Complete queryable event history

Same memory on web, desktop, mobile, and CLI

Weekly improvement cycle. Extract, approve, inject.

Built by people who needed it

“We built Clawtex because we kept hitting the same wall. Agents that forget. Now it works on every surface we use. The memory follows the work, not the device.”

Paul Evans, Founder -- Evans Studio

Pricing

Simple, fair pricing.

Free to start. Upgrade when you need more.

Free

£0/month
  • 1 agent
  • Connector access (all surfaces)
  • Structured state store
  • 14-day event history
  • Read-only dashboard
Recommended

Pro

£9/month
  • 3 agents
  • Connector access (all surfaces)
  • Unlimited event history
  • Full dashboard access
  • AI lesson extraction

Team

£24/month
  • Unlimited agents
  • Everything in Pro
  • Connector access (all surfaces)
  • Priority support

Stop re-explaining. Start compounding.

Free to start. One click to connect. Your agent gets better every week.