Claude Code Plugin MCP Server

UseMyContext for Claude Code

MIT license | usemycontext.ai | Docs | usemycontext SDK on npm

UseMyContext is the personal context layer for AI. You keep one profile of who you are and what you are working on, and any AI client reads it over MCP. This plugin brings that into the terminal: sign in once and Claude Code answers with your context already in the room - who you are, what this project is to you, what your documents say.

What it looks like

cd ~/code/acme-api
claude

> Who am I, and what is this project to me?

  This folder is mapped to your @work context. You're a backend engineer
  at Acme, this repo is the payments API you own, and your notes say the
  current focus is the v2 billing migration. Want me to pull the migration
  checklist from your files?

No pasting your background into every session. The context is already there, scoped to the folder you are in.

Install

/plugin marketplace add usemycontext/claude-code-plugin
/plugin install usemycontext@usemycontext

Then run /mcp, select usemycontext, and complete the browser sign-in.

You will need:

• A UseMyContext account. The free tier is enough to try this. No account yet? Create a free one at usemycontext.ai first.
• A current Claude Code (any version with /plugin support).

Sign-in is OAuth in the browser; no token is written to any file. This repo is also the plugin's marketplace, which is why the install is two commands. The second command is plugin@marketplace; both happen to be named usemycontext. Full reference docs, including the .umc format and the two connection modes, are in usemycontext/README.md.

Try it in 30 seconds

No signup needed. At the browser sign-in, use demo@usemycontext.ai with code 424242 (fixed - no email access needed). A shared, read-only demo account with a full profile, files, and a shared context. Then ask Claude Code "what do you know about me?". Docs: usemycontext.ai/docs

Disconnecting and revoking

You are never locked in:

• In Claude Code, run /mcp and disconnect usemycontext.
• Or revoke from the UseMyContext web app: the Connect page lists every AI connected right now, each with its own Disconnect button, plus a "Disconnect all".

Revoking is enforced on the server: the token dies immediately, everywhere. Nothing needs cleaning up locally.

What you get

Ask Claude Code things like "what do you know about me?", "summarize my notes on the Q3 launch", or "how many rows in expenses.csv are over 500?" - it reads your profile, searches and opens your files, answers from your documents with cited passages, and runs exact counts and sums over your spreadsheets. A /umc skill checks or changes which profile a folder reads, and a SessionStart hook points every new session at the right project automatically.

Under the hood: the UseMyContext MCP server at https://mcp.usemycontext.ai/mcp (Streamable HTTP, OAuth 2.1) with eight tools - profile, list_files, search_files, get_file, ask_docs, query_table, suggest_update, shared_context - all read-only against your files. One tool, suggest_update, can write: it files a pending suggestion that you review and approve in the web app. Nothing the AI does ever edits your profile or your files directly, and every read is logged to your audit trail.

One folder, one context

A work repo reads your @work profile. A side project reads your @personal one. Same terminal, different context, chosen by the folder you are in.

To bind a folder to a project, drop a .umc file in the repo root (usemycontext/.umc.example is a ready-made template):

projectId=p2
handle=@work

projectId is what scopes the reads. handle is an optional readable label. Find both in the web app on the project's page, or just ask the /umc skill to create the file for you. With no .umc, the folder reads your account's active profile.

What it can and cannot do

• Reads: your profile and the files you uploaded to UseMyContext, scoped to the connected or mapped project.
• Writes: nothing to your files. The one write tool, suggest_update, files a pending suggestion that you review and approve in the web app.
• Every read is logged to your audit trail.
• You can revoke access at any time, from /mcp or from the web app's Connect page.
• It does not change your Anthropic account, your Claude Code login, your billing, or which model runs.

Questions you probably have

What does the AI actually see? The profile you wrote and the files you uploaded to UseMyContext, scoped to the connected or mapped project. Nothing else.

Can it see my other projects? The server only ever serves projects you own. With the default account-wide sign-in, a folder reads its mapped project (or your active profile), and a request for any project that is not yours is refused. If you want hard isolation per repo, use a single-project sign-in: that token is scoped to one project and cannot read the others.

Does it phone home from my repo? No. The plugin reads only the local .umc marker to know which profile to use. It never uploads your code or files; the only thing that travels is Claude's tool calls to your UseMyContext account.

What does it cost? The plugin is free and MIT-licensed. UseMyContext has a free tier that covers trying all of this; paid tiers are at usemycontext.ai/pricing.

Repo layout (for contributors)

.
|-- .claude-plugin/
|   `-- marketplace.json      # the marketplace catalog (one plugin: usemycontext)
`-- usemycontext/             # the plugin
    |-- .claude-plugin/
    |   `-- plugin.json        # plugin manifest
    |-- .mcp.json              # registers the remote UseMyContext MCP server (OAuth)
    |-- skills/umc/SKILL.md    # the /umc skill (status + folder mapping)
    |-- hooks/hooks.json       # SessionStart hook
    |-- scripts/umc-session.sh # reads a folder's .umc marker
    |-- .umc.example           # example folder mapping
    `-- README.md              # reference docs for the plugin