# Architecture at a Glance

> **TL;DR** OCL has three layers. \
> \
> 1\. Memory Buckets store your context. \
> 2a. AI Context Flow lets you manage, share, and work with your context on any website or agent. \
> 2b. The Plurality MCP Server exposes your context to MCP-compatible AI agents and tools. \
> 3\. Your AI agents and tools consume the context. \
> \
> You own everything in the bottom layer (your memory buckets).

#### The four layers

```
                    ┌─────────────────────────────────┐
                    │  AI Agents                      │   The consumers
                    │  Claude · ChatGPT · Cursor      │
                    └────────────────┬────────────────┘
                                     │  MCP
   ┌─────────────────────────┐  ┌────▼─────────────────────┐
   │  AI Context Flow        │  │  Plurality MCP Server    │   The clients
   │  (Browser Extension)    │  │                          │
   │  capture · organize ·   │  │  exposes memory buckets  │
   │  permission · share     │  │  to AI agents            │
   └────────────┬────────────┘  └─────────────┬────────────┘
                │                             │
                │   read / write              │   read / write
                ▼                             ▼
   ┌──────────────────────────────────────────────────────┐
   │  Memory Buckets                                      │   The primitive
   │  user-owned · encrypted · portable                   │
   └──────────────────────────────────────────────────────┘
```

#### Layer 1 — Memory Buckets (the primitive)

A memory bucket is an encrypted, user-owned container of context. You can have multiple bucket for different parts of your life e.g. work, personal, health, a side project, and each one holds whatever context belongs to that slice of you.

#### Layer 2a — AI Context Flow (the product)

Memory buckets on their own are just storage. AI Context Flow is the product layer that makes them useful. It's a browser extension that helps you capture context from your activity, help you organize it into the right bucket, and lets you set permissions on what gets shared and with whom.

Think of it as the operating system for your context: capture, sort, permission, share.

→ Read more: [AI Context Flow](/ai-context-flow.md)

→ Download here: [AI Context Flow](https://chromewebstore.google.com/detail/ai-context-flow-use-your/cfegfckldnmbdnimjgfamhjnmjpcmgnf)

#### Layer 2b — The Plurality MCP Server (the bridge)

The Plurality MCP Server speaks the [Model Context Protocol](https://modelcontextprotocol.io), an open standard for connecting AI agents to data sources. By exposing your Memory Buckets through MCP, any compatible agent can request context from your profile and (with your permission) receive it in real time.

With MCP, you can also set it up as a context store for your programmatic agents to read and write from e.g. OpenClaw.

You don't paste your background into every new ChatGPT conversation or every new agent session. The agent asks; the MCP Server answers — under the rules you set.

→ Read more: [The Plurality MCP Server](/the-plurality-mcp-server.md)

#### Layer 3 — Your AI agents (the consumers)

The agents you already use. Claude (Desktop and Code), ChatGPT, Cursor, Windsurf, GitHub Copilot, LM Studio, Lovable, Replit, OpenClaw, and any other tool that speaks MCP. They see the context you choose to share, and nothing else.

→ Read more: [Connect Your Agents via MCP](/the-plurality-mcp-server/connect-your-agents-via-mcp.md)

#### How a request flows (the MCP Way)

When you ask Claude *"what was I working on last week?"* and have setup the MCP connection:

1. **Claude** receives your question and recognizes it needs context.
2. **Claude** queries the **Plurality MCP Server** through the standard protocol.
3. **The MCP Server** checks which bucket has the relevant context and retrieves it from the correct bucket. 
4. The context is returned to Claude, which uses it to answer.

#### How a request flows (the Browser Extension Way)

Alternatively, if you are on a browser AI Agent and not using MCP. 

When you ask ChatGPT website *"what was I working on last week?"* and then press "Optimize" button

1. Your Prompt is improved and the relevant context is fetched and added from the selected memory bucket into your prompt
2. ChatGPT gives you improved answers because it now got a better, more contextual prompt with all the right details<br>

> Note: The prompt optimization feature in AI Context Flow browser extension currently only works on the websites of ChatGPT, Claude, Gemini, Perplexity and Grok.

In both cases, you see a better, more fitting answer. The context never leaves your control without permission, and the answer is grounded in *your* history and is not a generic guess.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.plurality.network/overview/architecture-at-a-glance.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.