covate

Overview

covate is open-core developer learning. A free open-source MCP sidecar turns your AI-assisted code changes into instant quizzes and keeps persistent debugging memory across sessions, plus a cloud learning platform you sign into with GitHub — it keeps a record of your sessions. It plugs into AI coding assistants rather than replacing them.

Who it’s for

Developers who use AI coding assistants and want to actually retain what the AI does with them — turning passive acceptance into durable learning.

What it does

• Turns AI code changes into instant quizzes
• Persistent debugging memory across sessions
• MCP sidecar that plugs into AI coding assistants
• Open-core: free sidecar plus a cloud platform you sign into with GitHub

How it works

1. It quizzes you on the diff

   Covate is an MCP server that sits beside your AI coding assistant. When the assistant writes something, you can ask for a quiz on it — logic, security or performance implications — so accepting a change and understanding it stop being the same action.

2. And it remembers debugging

   The server quietly records how a bug was solved and retrieves it when a similar error appears again, giving the agent a project-specific debugging memory instead of solving the same problem twice.

3. Spaced repetition on your own code

   A hosted learning ledger at covate.org tracks what you have been quizzed on, so review lands on your own diffs rather than on generic exercises.

Built with

TypeScript, implementing the Model Context Protocol. Listed on the Glama MCP marketplace, with generated deep-dive documentation on DeepWiki and the source open on GitHub.

Current status

Live and actively developed — the most-starred of the eight after sunkit.

Setup

covate is a Python MCP server (3.11+) that installs from its GitHub repository — it is not published to npm or PyPI, so neither npx covate nor pip install covate will find it. The installer clones the repo to ~/covate, builds a virtualenv beside it, and prints the executable path you point your assistant at.

1. Install — macOS / Linux

   Detects uv, conda, or venv; clones to ~/covate; installs dependencies; prints the command path to use below.

   curl -fsSL https://raw.githubusercontent.com/SunflowersLwtech/covate/main/scripts/install.sh | bash

2. Install — Windows (PowerShell)

   irm https://raw.githubusercontent.com/SunflowersLwtech/covate/main/scripts/install.ps1 | iex

3. Claude Code — one command

   Use the path the installer printed. On a default install that is ~/covate/covate/bin/covate (Windows: %USERPROFILE%\covate\covate\Scripts\covate.exe).

   claude mcp add covate -- ~/covate/covate/bin/covate

4. Claude Code — config file (~/.claude.json)

   Equivalent to the command above if you would rather edit JSON. Transport is stdio, which is the default.

   {
     "mcpServers": {
       "covate": {
         "command": "~/covate/covate/bin/covate"
       }
     }
   }

5. Cursor — ~/.cursor/mcp.json

   Settings → MCP → Add Server writes the same entry. Windsurf takes an identical block in ~/.codeium/windsurf/mcp_config.json. Restart the editor afterwards.

   {
     "mcpServers": {
       "covate": {
         "command": "~/covate/covate/bin/covate"
       }
     }
   }

6. Optional — sync sessions to the learning ledger

   Sign in at covate.org with GitHub, reveal your sync token in the dashboard, then run this from a project directory. The MCP works fully without it.

   COVATE_SYNC_URL=https://covate.org COVATE_SYNC_TOKEN=<your token> ~/covate/covate/bin/python -m covate.platform_sync

Environment variables

• MCP_DEBUG default: false Optional. Debug logging; accepts true, 1, yes, or on.
• COVATE_SYNC_URL default: unset Optional. Set to https://covate.org only when pushing sessions to the hosted ledger.
• COVATE_SYNC_TOKEN default: unset Optional. Your per-user token from covate.org/dashboard → Sync token.

If it doesn’t work

Commands come from the covate repository — github.com/SunflowersLwtech/covate — and are current as of its README and scripts/install.sh. If the repo and this page ever disagree, the repo is right.

Pricing

Free.

• The MCP server is free and open
• The hosted learning ledger is free

FAQ