Use cases

The knowledge base
your agents can read.

Teams building with AI need notes that machines can navigate and humans can trust. NoteX gives agents an MCP server and scoped API keys, and gives you the sessions, prompts, and review flow to stay in charge.

The pitch

Your team runs on agents.
Give them notes they can read.

AI-powered teams hit the same wall: the knowledge the agent needs is locked in a tool it cannot see, or in formats that burn context. NoteX is built for the other way around. Notes are structured files and API resources, so Claude Code, Cursor, or your own agents work the knowledge base directly.

Say it as a loop: your AI works inside NoteX natively, and NoteX works inside your AI through MCP.

AI integration

Before we scope Search v3, what did we learn from the v2 beta? Check the wiki.

I searched the Atlas Labs workspace over MCP and read 6 pages. Key finding: semantic ranking won 7 of 9 test queries. Proposing:

Create page: Search v3 Scope DraftSources linked: spec, beta notes, sprint 14
Applied
Append open questions to Product Spec: Search v23 items for Maya Chen
Proposed

An agent researching the wiki over MCP. It proposes pages; a human applies them.

MCP

Ten tools,
one config block.

The NoteX MCP server exposes your knowledge base as 10 tools: search, read, create, update, append, trash and friends. Wire it into Claude Code, Claude Desktop, Cursor, or any MCP client and the agent treats your wiki the way it treats a codebase: something it can navigate, quote, and change.

Under it sits a Notion-style REST API, so if your team already scripted against Notion, the mental model (and most of the code) carries over.

The API and MCP server
🔌

Wiring NoteX into Claude Code

Team Wiki › AI setup · Omar Haddad

One MCP server entry, and the agent can search, read, create, update, append, and trash pages:

json
{
  "mcpServers": {
    "notex": {
      "command": "npx",
      "args": ["notex-mcp-server"],
      "env": { "NOTEX_API_KEY": "nx_…" }
    }
  }
}
🔑

Use a read-only key for research agents; give the support bot a key scoped to its own page tree.

Access control

Scoped keys,
not shared passwords.

Every agent gets its own nx_ API key with exactly the power it needs: read-only for research agents that should never touch a page, page-scoped for a support bot that lives in one tree, expiring keys for pipelines and contractors. Usage is logged per key, so you can see which agent read what, and revoke one without breaking the rest.

AgentIts nx_ key
Research agentRead-only, whole workspaceCan never write
Support botRead/write, scoped to Support › FAQ treePage-scoped
CI release notesRead/write, Releases tree, 90-day expiryExpires
Contractor's agentRead-only, one page tree, 2-week expiryShort-lived

One key per agent, least privilege by default. Revoke one, the others keep working.

Context economics

Smaller context,
smarter agents.

Agents pay for every token they read. A structured .notex or Markdown file carries the same content as a PDF at a fraction of the context: up to ~50x smaller in our internal benchmarks. That budget goes somewhere useful: agents read more pages per task, answer with more of your knowledge in view, and cost less per run.

Try the context calculator
Scanned PDF
≈ 180K tokens
Word document
≈ 82K tokens
NoteX file
≈ 3.6K tokens

Illustrative comparison from our internal benchmarks: one 40-page product spec in three formats.

The human side

Built for the people
running the agents.

🪟 Agent sessions window · desktop

Long-running agent work gets its own window, out of your writing surface. Watch Claude Code refactor the wiki without it camping in your editor.

🧰 AI customization · desktop

One system for agents, skills, prompts, instructions, hooks, and MCP servers. Your team's agent setup becomes configuration, not tribal knowledge.

📚 Prompt library & personas · web

The prompts that work get saved and shared with the team, and personas keep the AI sidebar answering in the voice each job needs. Leading models including Claude, one picker.

Everything AI in NoteX

Frequently asked questions

Which AI tools can connect to NoteX?

Any MCP client: Claude Code, Claude Desktop, Cursor, and your own agents. The MCP server exposes 10 tools (search, read, create, update, append, trash and more). Anything else can use the Notion-style REST API directly with an nx_ key.

How do I stop an agent from editing pages it should not touch?

Scope its key. nx_ keys are read-only or read/write, can be limited to a single page tree, and can expire. A research agent gets read-only on the workspace; a support bot gets read/write on its own tree and nothing else. Usage logging shows exactly what each key did. See API and MCP.

Why do agents do better with NoteX files than with PDFs?

Structure and size. A .notex or Markdown file is clean, ordered text, and in our internal benchmarks the same content can be up to ~50x smaller than a PDF. Smaller context means the agent reads more of your knowledge per task. Try the context calculator.

Does AI edit pages without asking?

No. On web, the AI sidebar proposes edits with a diff preview and you apply them with one click. Agents working over the API write only where their key allows, and every write is logged. On desktop, add-ons like Claude Code show you proposed changes before they land.

What about AI for teammates who do not run agents?

NoteX web has an AI sidebar with streaming chat, one-shot actions (summarize, translate, fix grammar, extract action items), autocomplete ghost text, and a shared prompt library with personas, across leading models including Claude. See AI integration.

Your notes deserve a better engine.

A Notion-style block editor on the engine behind the world's most popular code editor. Notes stay files you own, extensions add what you need, and AI plugs in natively. Free to start, on web and desktop.

Get NoteX free Book a demo

Free to start · Your files, your disk · Windows desktop & web