Glossary
Definition

Model Context Protocol

An open protocol developed by Anthropic that enables AI assistants to securely connect to external data sources and tools.

Full definition

The Model Context Protocol (MCP) is an open standard developed by Anthropic that defines how AI assistants communicate with external data sources, tools, and services. MCP enables AI models like Claude to move beyond their training data and interact with live, real-world information — including files, databases, APIs, and personal knowledge graphs — in a structured, secure, and standardised way.

History & origin

Anthropic introduced MCP in late 2024 as a response to the fragmentation of AI integrations. Prior to MCP, every AI assistant required bespoke integrations for each external tool — a Notion connector for Notion, a GitHub connector for GitHub, and so on, each built to a different specification. MCP established a universal protocol: any tool that implements the MCP server specification can be connected to any MCP-compatible AI client, including Claude Desktop. The protocol is open-source and has attracted rapid adoption from developer tools, productivity apps, and data infrastructure companies.

Key concepts

MCP Servers

An MCP server is a lightweight program that exposes a data source or tool to MCP-compatible AI clients. An MCP server for a note-taking app, for example, would expose functions like 'search notes', 'get note by ID', and 'list recent notes'. Brinn's MCP server exposes your personal knowledge graph to Claude.

MCP Clients

An MCP client is an AI application that can connect to MCP servers to retrieve context. Claude Desktop is the primary MCP client, allowing users to connect any number of MCP servers — giving Claude access to their files, code, notes, databases, and more simultaneously.

Context injection

When you ask Claude a question, the MCP client queries connected servers for relevant context before generating a response. If you ask 'What did I decide about the marketing strategy last month?', Claude queries your connected knowledge graph (via MCP) for relevant notes, then uses that context to answer accurately from your own data.

Security model

MCP is designed with a local-first security model. MCP servers typically run on your local machine or within your trusted infrastructure. Data flows from your server to the AI client on demand — nothing is sent proactively or without your prompting. Authentication is managed via API keys.

How Brinn applies this

Brinn ships an official MCP connector that gives Claude Desktop real-time access to your personal knowledge graph. After a one-time setup (add a config entry to Claude Desktop's settings file), Claude can query your Brinn notes, entities, and graph connections to answer questions about your own knowledge and history. The connector is available to all Brinn users at no additional cost.

Try it free

Frequently asked questions

What AI assistants support MCP?
Claude Desktop is the primary MCP client. Anthropic built MCP as an open standard, and other AI tools are beginning to adopt it. Check the MCP specification documentation for an up-to-date list of compatible clients.
Is MCP secure?
Yes. MCP servers run locally on your machine (or in your own infrastructure), and data is only retrieved on demand when you ask the AI a question. The Brinn MCP connector uses API key authentication and only shares the notes and entities you've stored in Brinn — nothing else.
How is MCP different from ChatGPT plugins?
ChatGPT plugins were a proprietary integration system limited to OpenAI's ecosystem. MCP is an open standard: any AI client can connect to any MCP server without platform-specific integration work. MCP also emphasises local-first security, whereas plugins operated via cloud API calls.
How do I connect Brinn to Claude via MCP?
Install Claude Desktop, generate an API key from Brinn's Settings → Integrations page, then add a Brinn server entry to Claude Desktop's configuration file. Full step-by-step instructions are on Brinn's /mcp page and /developers documentation.