Not another memory MCP. A knowledge system that tends itself

Notes and documents become a linked knowledge graph with hybrid search and page-level provenance, readable and maintainable by any MCP-compatible AI app via OAuth 2.1. Reasoning runs on your model. Hosted in the EU. Everything exports as Markdown, any time.

Try free for 14 days

From €2.99/month · card required · works on every Claude plan, Free included

One URL. Any AI app

Claude Code · every Claude plan

claude mcp add --transport http --scope user memexa https://brain.memexa.ai/mcp

Claude (claude.ai, desktop) · every plan, Free allows one custom connector

Settings → Connectors → Add custom connector → https://brain.memexa.ai/mcp

ChatGPT · tested on Free and Plus

chatgpt.com/plugins → Add → Add custom MCP server → name, server URL https://brain.memexa.ai/mcp, OAuth → Create as plugin

Mistral Vibe · free plan included, EU

Add a custom MCP connector → https://brain.memexa.ai/mcp

Cursor and other MCP tools

mcp.json → { "mcpServers": { "memexa": { "url": "https://brain.memexa.ai/mcp" } } }

Sign in once in the browser (OAuth 2.1), no tokens to copy. The endpoint is the same for every AI app. Plans and paths as of October 2026; menu names in the AI apps can change. Step by step: how to connect each AI app.

What "not a scratchpad" means

Document pipeline

PDFs are OCR'd, chunked, embedded and distilled into linked notes, so search hits carry page-level provenance.

Typed graph edges

Notes, people, projects and sources are linked with typed relations, not a flat tag soup.

Auto-linking

New captures are connected to existing knowledge automatically. No manual gardening.

Weekly digest

A summary of what changed, what connected, and what's left open. The graph tends itself.

Architecture in one line: your model reasons, Memexa is the substrate.

Against thin memory stores

Memexa Typical memory stores
Works across AI apps Any MCP-compatible AI app Single vendor, or plugins
Keeps itself organized Automatically, from day one: no rules to write, no usage credits Manual upkeep, or agents that run on rules you write
Ships its own model No, bring your own Bundled (you pay twice)
See & edit every memory Full notes UI + graph Opaque store or settings list
Export One click, complete: Markdown, JSON, original files Varies; often JSON dumps
Hosting EU, GDPR US by default

Based on publicly documented behavior, checked October 30, 2026. Some tools now tend themselves too: Basic Memory 1.0 added beta agents that link notes and merge duplicates, on instructions you write and billed via usage credits (as of November 7, 2026). Memexa tends its memory automatically from day one, without rules to write or credits to buy. For how Claude's own memory layers work and where they stop, see the Claude memory guide.

Your data stays yours

Hosted in the EU under the GDPR. Every note is Markdown you can read, edit and export; everything else exports as JSON, your original files as a ZIP. The graph is an index, not a lock-in.

Pricing is deliberately boring: Basic €2.99/mo or €29.90/yr (all AI apps, 10 document pages/mo) · Pro €9/mo or €90/yr · Power €19/mo or €190/yr · every plan starts with a 14-day trial (card required, reminder before it ends), cancel any time.

Try free for 14 days

From €2.99/month · card required · reminder before the trial ends