How to Load Personal Context into Claude, ChatGPT and Cursor via MCP
Step-by-step MCP setup for Claude, ChatGPT, Cursor and Claude Code, plus what to do in tools that don't support MCP.

Every AI session starts from zero: you re-explain your role to Claude, then to Cursor, then to ChatGPT. MCP fixes that by letting each tool pull your personal context from one place when it needs it, instead of you pasting it in.
Below is the setup for each tool, how to check it worked, what each connection can see, and what to do where MCP isn't available.
What does loading personal context via MCP mean?
It means the AI tool fetches your context from a server when it needs it, instead of you pasting it into the chat. The Model Context Protocol is the open standard for that connection. Anthropic released it in November 2024, and by December 2025 there were more than 10,000 published MCP servers, with Claude, ChatGPT, Cursor, VS Code and Gemini all supporting it (Linux Foundation, 2025).
There are two ways context reaches a model. Pull: the tool requests it over MCP, so it's current every time. Push: you paste a file or fill in custom instructions, and it's only as fresh as your last edit. Pull is the one worth setting up wherever the tool supports it.
Unabyss runs a remote MCP server at https://mcp.unabyss.com that exposes tools, not static files. The ones agents use most:
whoamireturns your identity summary, the cheap baseline read at the start of a session.querylooks up anything about you, your work or your preferences in plain language.storesaves a durable fact back to your context.agentic_queryruns a deeper, multi-step search when one lookup isn't enough.export_listandexport_readfetch prebuilt Markdown bundles of your context.
The full reference, including OAuth and error codes, is in the MCP API docs.
What do you need before you start?
You need a Unabyss account with at least two connected apps, and an AI tool that supports remote MCP. Sign up, connect sources such as Gmail, Google Calendar, Notion, Slack, GitHub or Linear, and Unabyss builds your context from them (how it works). The first-connection setup over MCP won't move past its apps step until two are connected.
How do you connect Claude?
Use the official Unabyss connector from Claude's directory; there's no URL to paste.
- Open the Unabyss connector page and click through to Claude's connector directory.
- Click Connect and sign in with Unabyss when prompted.
- Start a new chat and send "Let's set up Unabyss". The setup walks you through the rest.
Remote connectors are tied to your Claude account and connect from Anthropic's cloud, which is how they reach claude.ai, Claude Desktop and the mobile apps (Claude Help Center). You don't need to edit claude_desktop_config.json; that file is for local MCP servers running on your machine.
If you'd rather add it by hand on a Pro or Max plan, go to Customize > Connectors, click +, then Add custom connector, and enter https://mcp.unabyss.com. Custom connectors work on every Claude plan, but Free users are limited to one.
How do you connect ChatGPT?
Add the official Unabyss plugin; no Developer mode, no server URL.
- Open the ChatGPT setup page and go to the Unabyss plugin in ChatGPT.
- Add the plugin and sign in with Unabyss.
- Send "Let's set up Unabyss" in a new chat.
ChatGPT also lets you add any MCP server yourself: on the web, go to chatgpt.com/plugins, click the plus button, choose Add custom MCP server, enter the URL, pick OAuth, and select Create as a plugin. You then install it and call it with @ in the prompt box (OpenAI docs). For Unabyss the official plugin does the same in one step. Codex uses the same plugin, opened from the Codex desktop app.
How do you connect Cursor?
Use the one-click install, or add the server to mcp.json.
- In the Unabyss app, open the MCP page at app.unabyss.com/mcp and click Connect on the Cursor card. Cursor opens and asks to install the server.
- Sign in when Cursor starts the OAuth flow.
To do it by hand, add this to ~/.cursor/mcp.json for every project, or to .cursor/mcp.json for one project:
{
"mcpServers": {
"unabyss": {
"url": "https://mcp.unabyss.com"
}
}
}
Cursor supports OAuth for remote servers, and you manage installed servers from its Customize page (Cursor docs). If you'd rather use a static token, generate one on the Unabyss MCP page and add "headers": { "Authorization": "Bearer <token>" } to the block.
How do you connect Claude Code?
Add the server from your terminal, then sign in from inside a session:
claude mcp add --transport http unabyss https://mcp.unabyss.com
Run /mcp in Claude Code and follow the browser login (Claude Code docs). Add --scope user to make it available in every project. To use a static token instead of OAuth, append --header "Authorization: Bearer <token>".
What about other AI tools?
Most MCP hosts take the same URL. VS Code has a one-click install on the Unabyss MCP page. Grok takes it as a custom connector at grok.com/connectors. Hermes Agent uses hermes mcp add unabyss --url "https://mcp.unabyss.com" --auth oauth, then hermes mcp login unabyss. For OpenClaw, Gemini CLI and Antigravity, the Get Token & Prompt button on the MCP page gives you a bearer token and the config to paste.
Static tokens start with unby_mcp_ and are shown once, so store them in a password manager. You can revoke any token or OAuth connection from the same page.
How do you check it's working?
Ask the tool "What MCP tools are available?" You should see Unabyss tools such as whoami, query and store. Then ask something only your context can answer, such as "What am I working on this week?" If the answer names your real projects, it's working.
A generic answer usually means the tool didn't call Unabyss. Tell it to call whoami at the start of each session and query whenever an answer depends on who you are; the setup flow ends with a system prompt for your tool that covers this.
What can each connection see?
Each connection gets its own permissions, set when you create or edit a token, or on the OAuth consent screen:
- Exclude private info drops personal-life material from results.
- Exclude company confidential drops company-confidential material.
- Excluded apps blocks chosen integrations, so their data never enters a
queryoragentic_queryanswer.
Two connections on the same account can return different answers to the same question. That's the point: a coding agent doesn't need your personal email. On usage, whoami is free, while query, store and agentic_query count against your Unabyss credits (MCP API docs). Whether connecting a tool is safe in general is covered in Is MCP Safe?
What if a tool doesn't support MCP?
Export your context as a Markdown file and paste or upload it. Unabyss offers more than 20 one-click exports, each formatted for a specific tool, plus custom exports. They work anywhere that accepts text: a Claude Project, ChatGPT custom instructions, Notion AI or a local model.
| Method | Setup | Stays current | Best for |
|---|---|---|---|
| MCP | One connection per tool | Yes, pulled on each request | Claude, ChatGPT, Cursor, Claude Code |
| Markdown export | None | No, re-export after changes | Tools without MCP, one-off sessions, sharing with a collaborator |
The trade-off is freshness. An export is a snapshot: start a new project or change roles and the file doesn't know. Use MCP wherever it's supported and exports everywhere else.
Is MCP worth setting up for personal context?
Yes, if you use more than one AI tool. You connect once per tool, and from then on every session starts with what's true about you now. In an August 2026 Unabyss benchmark in Claude Code, adding Unabyss made the same 20 tasks 36% cheaper and 38% faster than connected tools alone, with 60% fewer follow-ups (What Structured AI Context Is Worth).
If you're still choosing a tool, compare the options in Best AI Memory Tools. If you haven't built your context yet, start with How to Build Personal Context for AI. Or connect your apps to Unabyss and set up your first MCP connection today.