Ask Claude a question and you get a smart, technically correct answer. Ask it a question about something you have been researching for six months, and you get the same smart, technically correct answer.
The problem is not capability. Claude knows a lot. The problem is that it does not know what you know.
Every conversation starts blank. No memory of the posts you have saved. No awareness of the documents you have written. No access to the decisions you have made or the topics you have been tracking for months. Generic input, generic output.
The Model Context Protocol (MCP) is the practical fix for this. It is an open standard created by Anthropic and now adopted by every major AI provider, including OpenAI, Google, and Microsoft. It lets you connect external data sources to AI tools once, and Claude can pull from them mid-conversation whenever relevant. No copy-pasting. No re-explaining context you have already explained.
The question is which data sources are actually worth connecting.
Not all personal data is equally useful as AI context. Some sources are dense with signal. Others are noise. This guide covers the five that are worth the setup time, starting with the most overlooked.
- Bookmarks: web pages plus X, Reddit and LinkedIn saves, via ContextBolt. Free to start, and it fills itself from the day you install it.
- Notes and documents: Obsidian, Notion, Google Docs.
- Calendar.
- Email.
- Project files: local filesystem.
- Each adds a layer of personal context that turns generic AI answers into specific ones.
- Start with your bookmarks, because that library only exists from the day you start capturing, then add the source that holds your most active knowledge, one at a time.
1. Your bookmarks
This is the one most guides skip entirely. It is, in my view, the single most underrated personal data source you can add to Claude.
Every time you bookmark a tweet, save a Reddit post, or save a LinkedIn article, you make an active decision. You are saying this is worth keeping. That signal is genuinely rare in a world where most content is passively scrolled past.
Your notes contain things you wrote. Your calendar contains appointments. But your bookmarks contain things you decided were worth saving from the stream of content you consumed. Over months and years, that collection becomes a curated knowledge base built around your actual interests and work areas. We make the longer argument for this framing in Your Bookmarks Are Personal Documentation.
The problem is accessing it. X added bookmark search in July 2024, but users report it is unreliable and frequently misses exact keyword matches. Reddit has no search at all for saved posts and shows only your most recent 1,000, so older saves drop off the reachable list. LinkedIn has no search for saved content whatsoever.
So the data exists. It is locked away.
ContextBolt fixes this. Install the Chrome extension and what you bookmark on X, save on Reddit or save on LinkedIn is captured as you make it, content included, with no resaving. Any web page saves with one click, article text included. One library, on your device.
Every captured bookmark is AI-tagged with a main topic and specific keywords on arrival, and clustered into topics. You get topic filtering in the extension, instant on-device search across your collection, and semantic search by meaning through the Pro MCP endpoint.
Pro users ($6 a month, month to month) get a personal MCP endpoint. Through it, Claude can search your bookmarks by meaning, list your topic clusters, pull the bookmarks in one topic, fetch your most recent saves, and save new bookmarks from the chat.
Connect that endpoint to Claude Desktop or Claude Code and you can ask “what have I saved about pricing strategy?” and get an answer drawn from the specific content you curated over months, not from what the internet generally says about pricing.
The difference between those two answers is significant.
2. Your notes and documents
This is the source most people think of first, and rightly so. Your notes hold your thinking. Your documents hold your decisions, your research, and your processes.
If you use Notion, Notion’s official MCP server is available through Claude Desktop’s built-in Connectors menu. No JSON editing, no terminal. Open Settings in Claude Desktop, go to Connectors, add Notion, complete the OAuth flow. Under two minutes.
Once connected, Claude can search your pages and databases, retrieve specific documents, and read content from your workspace. When you ask about a project you have been working on, it can pull the relevant Notion pages before answering rather than asking you to describe what you have already written.
For Obsidian users, a community MCP server exists for local vaults. For Google Drive, the Connectors menu handles that too. For local files on your machine, the Filesystem MCP server ships bundled with Claude Desktop and gives Claude read access to whichever folders you specify.
The key question to ask yourself is where your most important active working knowledge actually lives. Connect that source, not the neatest-looking one.
3. Your calendar
Knowing what you know is one layer of context. Knowing when you are is another.
A calendar-connected Claude can give you answers that account for your actual schedule. “How long will this realistically take?” becomes answerable against real availability rather than in a vacuum. “What should I prepare for tomorrow?” becomes a question it can address from your actual agenda rather than a generic template.
Claude Desktop’s Connectors menu includes Google Calendar as a built-in option. Setup takes under two minutes. You authorize read access, pick which calendars to include, and Claude can see your schedule from that point forward.
The most useful cases are meeting preparation and workload planning. Ask Claude to help you prepare for a meeting it can see on your calendar and it has something concrete to work with. Ask it whether you have bandwidth for a new project this week and it answers from actual data rather than asking you to describe your schedule.
Read-only access is enough for most of what makes this useful. Claude does not need to create calendar events to give better advice.
4. Your email inbox
Email is where real decisions live. Your inbox contains the actual conversations behind the things you are working on, the commitments you have made, and the threads that need following up.
Claude connected to Gmail can read threads, summarize long email chains, and draft replies in context. Not “draft a professional follow-up email about a project review” but “draft a follow-up to the conversation with Sarah about the Q3 budget discussion from last Tuesday.”
The Connectors menu in Claude Desktop handles Gmail alongside Google Calendar. The same OAuth flow, the same setup time.
One limitation is worth knowing upfront. Email is high volume and often low signal. Connecting your inbox adds a large amount of data, some of which is noise. The biggest wins come from specific threads where you need full context before drafting a reply or making a decision. A focused “summarize this thread” request outperforms “tell me about my emails this week” for most people.
5. Your project files
If you write code, your codebase is the highest-signal data source you own. Claude connected to a GitHub repository can read the actual code before suggesting changes, review pull requests with full context, and give advice grounded in what the codebase actually contains rather than generic best practices.
If you do not write code, local files still add clear value. The Filesystem MCP server lets you point Claude at a folder of documents, reports, or spreadsheets. When you ask it to help edit or analyze something, it reads the actual file rather than working from whatever you paste into the chat.
Setup for local files involves a small JSON edit, adding the Filesystem server to your Claude Desktop configuration file and specifying which paths to allow. Claude can then read anything in those paths. No coding required, and the Claude Desktop setup guide walks through it in under five minutes.
How the five sources compare
| Data source | How to connect | Best use case | Native in Claude Desktop? |
|---|---|---|---|
| Bookmarks | ContextBolt Pro MCP endpoint | Research recall, topic exploration | Custom connector, one URL, on ContextBolt Pro ($6/month) |
| Notes & documents | Connectors (Notion, Drive) or MCP server (Obsidian, local files) | Knowledge work, project context | Yes (Notion, Google Drive) |
| Calendar | Connectors menu (Google Calendar) | Meeting prep, workload planning | Yes |
| Connectors menu (Gmail) | Thread context, drafting replies | Yes | |
| Project files | Filesystem MCP or GitHub MCP | Code review, document editing | Filesystem: yes. GitHub: separate install. |
How to pick which personal data source to start with
Do not connect all five at once. More sources add noise before they add signal. Claude has to decide which source is relevant to each question, and that decision gets harder when everything is connected simultaneously.
The right approach is sequential, and the first step is the same for everyone.
Start with your bookmarks. Four of the five sources already exist and will be waiting whenever you connect them. Your notes are in Notion today and will still be there next month. A bookmark library is different, because it only holds what was captured after the extension went in, and every week before that is saving your agent will never see. Installing ContextBolt takes a minute, Basic is free, and from then on the collection builds itself while you set up everything else. A collection built over four weeks is meaningfully more useful than one you start building after four weeks of delay. Add the MCP endpoint on Pro when you are ready to ask Claude about it.
Then add the source that holds your most active, highest-stakes knowledge. If you work in code, that is filesystem or GitHub. If you are a knowledge worker, it is Notion or Google Drive. Connect it and use it for two weeks before adding anything else.
Calendar and email add the most value for specific tasks, such as scheduling, meeting preparation, and drafting context-aware replies. They are not as universally high-signal as notes and bookmarks. Add them once your first two sources are working reliably.
Why personal data sources matter for AI agents
Context engineering is the skill of shaping what an AI agent knows before it answers your question. The right context turns a generically capable model into a tool that feels like it actually understands your work.
None of these five sources requires technical knowledge to set up. The most involved is bookmarks, and that takes less than five minutes with ContextBolt installed and the MCP endpoint configured.
The difference in answer quality is real. Generic Claude is impressive but impersonal. It draws on vast training data and gives you statistically common answers. Context-connected Claude draws on your documents, your schedule, your research. The answers reflect your specific situation rather than the average situation.
That is the version worth building toward. One source at a time.
ContextBolt is a Chrome extension that saves any web page with its article text in one click and captures your X, Reddit and LinkedIn saves automatically as you make them, all into one library stored on your device. Basic is free, with no account and no card. It is capped at 150 bookmarks, and includes AI tagging, topic clusters and instant local search. Pro at $6 a month, month to month, removes the cap and adds cloud sync and the personal MCP endpoint, so everything you have kept is searchable by meaning inside every Claude conversation. Install it first, because the library only starts filling the day you do.