What is the Model Context Protocol (MCP)?

Before MCP, every connection between an AI assistant and an app was a one-off. A task manager that wanted to work inside one chat product wrote a plugin for it, then another for the next product, and each assistant had a different way to describe what a tool did and how to call it. Only large vendors could afford to keep up.

MCP matters because it standardizes the conversation. An app runs an MCP server that describes its tools, such as “create a task” or “find tasks due today,” in a common format, along with the inputs each tool accepts. Any MCP client, Claude being the first, can read that description, show the tools to the model and call them when the user asks. Authentication is handled with standard OAuth, so you sign in once and the assistant acts as you.

For people who are not developers, the practical change is that an AI chat stops being a place where you talk about your work and becomes a place where the work gets done. You can ask for a summary of what is due and have the same assistant reschedule it, without switching apps. That cuts context switching and the decision fatigue of choosing what to do next.

Using an existing MCP server takes a few minutes; building one is a developer task. For the everyday case:

• Find the server URL. Apps that support MCP publish a hosted endpoint, often ending in /mcp, on their integrations page.
• Add it as a custom connector. In Claude, open connectors in settings, choose to add a custom connector and paste the URL.
• Sign in through OAuth. The connector redirects you to the app to approve access, then back to the chat. No API keys are copied around.
• Check which tools appeared. The client lists the actions the server exposes. Read them once so you know what the assistant can and cannot do.
• Ask in plain language. “Move everything due Friday to Monday” or “what is waiting on me?” The client picks the right tool and shows you the call before or after it runs, depending on the client.
• Review the results in the app. The first few times, open the app and confirm the changes look right.
• Disconnect what you stop using. Connectors can be removed from the same settings screen, which revokes the token.

For developers, the specification defines tools, resources and prompts, and there are SDKs for building servers in several languages.

Fluidwave publishes a hosted MCP server at https://fluidwave.com/api/mcp. Add it to Claude as a custom connector, sign in with your Fluidwave account through OAuth, and Claude gains 23 tools for your workspace: create, update, complete, cancel, delete and find tasks, singly or in bulk, and create, update and delete lists, schedules, categories and contexts. In the words of the setup screen, you’ll be able to complete tasks, get tasks, update tasks, categories, lists, and more from within Claude.

Because Fluidwave already uses Claude as its only language model, the experience inside the app and inside Claude is consistent. A task created from a chat carries the same fields, including urgency, duration and dates, and auto-prioritization orders it alongside everything else. Every change made through the connector is a normal Fluidwave mutation, so it syncs in real time to your other devices and can be undone with Cmd+Z in the app.

The connector is scoped to tasks and organizers; email, calendar and contacts are not exposed through it. If you want to build your own integration instead, the same OAuth flow backs a public REST API with an OpenAPI spec. Setup steps and screenshots are on the Claude MCP integration page, and the in-app assistant offers the same tools plus email, calendar and automation actions without leaving Fluidwave.