Skip to main content
Glama

Covate

English | 简体中文 | 繁體中文

A context-aware Model Context Protocol (MCP) server that acts as a learning sidecar for AI coding assistants. It helps developers learn from AI-generated code changes through interactive quizzes and provides agents with a persistent project-specific debugging memory.

License: MIT Python 3.11+ MCP Standard Docker Glama MCP DeepWiki


🌐 Resources

Resource

Description

Glama MCP Marketplace

Official MCP server listing with installation guides

DeepWiki Documentation

AI-generated deep analysis of the codebase

GitHub Repository

Source code, issues, and contributions


Related MCP server: Learning Assistant MCP Server

🚀 Why Use This?

For

Benefit

Developers

Don't just accept AI code—understand it. Request a quiz to verify your grasp of the logic, security, or performance implications.

AI Agents

Stop solving the same bug twice. The server quietly records debugging solutions and retrieves them automatically when similar errors occur.


☁️ Covate learning ledger (free)

The MCP server in this repo is free and open-source (MIT) — run it locally, no account required. The optional hosted learning ledger is free too: sign in with GitHub, then run COVATE_SYNC_URL=https://covate.org COVATE_SYNC_TOKEN=<your token> python -m covate.platform_sync from a project to push your local sessions up and review them in a browser:

  • ☁️ Sync your learning sessions from any machine

  • 📖 Every synced session, newest first, with its score

  • 📊 Totals — sessions, questions, correct answers, running accuracy

  • 🎯 The topics you answer worst, ranked

  • 🔑 Your sync token — reveal or rotate it whenever you want

Not built yet, so not promised: progress-over-time charts, spaced-repetition study plans, team accounts. There is no paid tier and nothing to buy — the MCP works fully without the ledger, and the ledger costs nothing.


🔁 Spaced review (local, offline)

covate review keeps a small local ledger of concepts and when to ask about them again. It never talks to the network or to a model; your own agent writes the questions and publishes the page (for example with the FoldRender command line).

covate review page questions.json -o page.html   # questions -> a quiz page (answers marked)
foldrender publish page.html                      # your agent publishes it
foldrender answers <page> --json | covate review record   # score the first answers
covate review due                                 # what to ask about again today
covate review status                              # every concept + first-answer accuracy

Only the first pick on each question counts, so changing an answer after seeing the explanation cannot raise the score. (FoldRender shows the explanation during its few-second undo window, so an answer you take back inside that window does not count; one you leave alone does.) A new concept comes back after 1 day, then 7, then 30 days; a wrong answer sends it back to the 1-day step; passing all three means mastered. The ledger lives in the global Covate storage (override with COVATE_REVIEW_LEDGER). questions.json is {"title": ..., "questions": [{"concept", "question", "options": [...], "correct": <0-based index>, "explanation"?}]}.

In Claude Code, add the marketplace, then install covate for /covate:review and /covate:review-record, which walk your agent through the steps above (no hooks, no server, nothing runs unless you call it), and covate-mcp for the MCP server:

/plugin marketplace add SunflowersLwtech/covate
/plugin install covate@covate
/plugin install covate-mcp@covate

covate is the one listed in Anthropic's plugin directory; covate-mcp is only in this marketplace, because the directory accepts servers that run on your computer only from organizations it has verified.

Installed covate@covate before 12 November 2026? It no longer includes the MCP server, and updating drops it without a message: run /plugin install covate-mcp@covate to keep it. Its tools are now named under plugin:covate-mcp:covate, so add any saved permission rules again.


📦 Available Tools

Tool

Type

Description

learning_session

🎓 Interactive

Opens a WebUI quiz based on recent code changes. Blocks until user completes learning.

debug_search

🔍 Silent RAG

Searches project debug history for relevant past solutions. Auto-triggered on errors.

debug_record

📝 Silent

Records debugging experiences to project knowledge base. Auto-triggered after fixes.

term_get

📚 Reference

Fetches programming terms/concepts. Tracks shown terms to avoid repetition.

Tool Details

Trigger: User explicitly requests (e.g., "Quiz me", "Test my understanding")

Parameters:

Parameter

Type

Default

Description

project_directory

string

"."

Project directory path

summary

string

—

Structured summary of Agent's actions

reasoning

object

null

5-Why reasoning (goal, trigger, mechanism, alternatives, risks)

quizzes

array

auto-generated

3 quiz questions with options, answer, explanation

focus_areas

array

["logic"]

Focus areas: logic, security, performance, architecture, syntax

timeout

int

600

Timeout in seconds (60-7200)

Returns: {"status": "completed", "action": "HALT_GENERATION"}

Trigger: Auto-called when encountering errors (silent, no UI)

Parameters:

Parameter

Type

Default

Description

query

string

—

Error message or description to search

project_directory

string

"."

Project directory path

error_type

string

null

Filter by error type (e.g., ImportError)

tags

array

null

Filter by tags

limit

int

5

Maximum results (1-20)

Returns: {"results": [...], "count": N}

Trigger: Auto-called after fixing bugs (silent, background)

Parameters:

Parameter

Type

Default

Description

context

object

—

Error context: {error_type, error_message, file, line}

cause

string

—

Root cause analysis

solution

string

—

Solution that worked

project_directory

string

"."

Project directory path

tags

array

null

Tags for categorization

Returns: {"ok": true, "id": "..."}

Available Domains: programming_basics, data_structures, algorithms, software_design, web_development, version_control, testing, security, databases, devops

Parameters:

Parameter

Type

Default

Description

project_directory

string

"."

Project directory path

count

int

3

Number of terms (1-5)

domain

string

null

Filter by domain

Returns: {"terms": [...], "count": N, "remaining": N}


🛠️ Installation

curl -fsSL https://raw.githubusercontent.com/SunflowersLwtech/covate/main/scripts/install.sh | bash
irm https://raw.githubusercontent.com/SunflowersLwtech/covate/main/scripts/install.ps1 | iex

The installer will:

  1. Auto-detect your Python environment (uv → conda → venv)

  2. Clone the repository to ~/covate

  3. Create virtual environment and install dependencies

  4. Print the exact command to configure your IDE

Manual Installation

Prerequisites: Python 3.11+ or uv

# 1. Clone the repository
git clone https://github.com/SunflowersLwtech/covate.git
cd covate

# 2. Create virtual environment and install
# Using uv (recommended)
uv venv --python 3.11 covate
source covate/bin/activate          # macOS/Linux
# covate\Scripts\activate           # Windows
uv pip install -e '.[dev]'

# Or using standard venv
python -m venv covate
source covate/bin/activate           # macOS/Linux
# covate\Scripts\activate            # Windows
pip install -e '.[dev]'

Docker Installation

Prerequisites: Docker installed on your system

# 1. Pull from Docker Hub
docker pull sunflowerslwtech/covate:latest

# Or build locally
git clone https://github.com/SunflowersLwtech/covate.git
cd covate
docker build -t covate .

# 2. Run with Docker
docker run -i covate

# 3. Or use Docker Compose
docker-compose up -d

For detailed Docker usage, persistent storage, and Claude Desktop integration, see DOCKER.md.


⚙️ IDE Configuration

Claude Code (CLI) — One Command Setup

After installation, configure your AI coding IDE to use this MCP server.

Claude Code

Option 1: CLI (Recommended)

# macOS / Linux
claude mcp add covate -- ~/covate/covate/bin/covate

# Windows
claude mcp add covate -- %USERPROFILE%\covate\covate\Scripts\covate.exe

Option 2: Config File

Add to ~/.claude.json:

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

For Windows:

{
  "mcpServers": {
    "covate": {
      "command": "C:\\Users\\YourName\\covate\\covate\\Scripts\\covate.exe"
    }
  }
}

Example paths:

  • Unix (uv): ~/covate/covate/bin/covate

  • Windows (uv): C:\\Users\\YourName\\covate\\covate\\Scripts\\covate.exe

  • Windows (conda): C:\\Users\\YourName\\anaconda3\\envs\\covate\\Scripts\\covate.exe

Path breakdown (Unix example):

  • ~/covate → repository directory

  • covate → virtual environment directory created by uv/venv

  • bin/covate → executable

Cursor

Add to Cursor MCP settings (Settings → MCP → Add Server):

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

For Windows:

{
  "covate": {
    "command": "C:\\Users\\YourName\\covate\\covate\\Scripts\\covate.exe"
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

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

Docker Configuration

To use Docker with any MCP-compatible IDE:

{
  "mcpServers": {
    "covate": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-v",
        "/path/to/your/project:/workspace",
        "-w",
        "/workspace",
        "covate"
      ]
    }
  }
}

See DOCKER.md for detailed Docker configuration examples for Claude Desktop, Cursor, and other IDEs.

Other IDEs

For any MCP-compatible IDE, use these settings:

  • Command: <install-path>/covate/bin/covate (or covate\Scripts\covate.exe on Windows)

  • Transport: stdio

After configuration, restart your IDE.

Usage

Available Tools

Tool

Trigger

For

Returns

learning_session

User explicit request

User

{status, action} - minimal

debug_search

Automatic (on error)

Agent

Compact summaries

debug_record

Automatic (after fix)

Agent

{ok, id} - minimal

For Users: Learning Session

Say to your AI assistant:

  • "Quiz me on this change"

  • "Test my understanding"

  • "Help me learn about what you did"

The agent will create an interactive learning card and wait until you complete it.

Note: Quiz scores are saved locally for your self-tracking but are NOT returned to the agent - this keeps the context clean.

For Agents: Debug Tools

The debug tools work silently in the background:

  • Search first: When encountering errors, agent searches past solutions

  • Record after: When fixing errors, agent records the solution

  • Progressive disclosure: Returns compact summaries, not full records

  • Fast lookups: Uses inverted index for keyword-based searches

Updating

The remote update script automatically detects your installation and works with any path format (including Chinese/non-ASCII paths):

curl -fsSL https://raw.githubusercontent.com/SunflowersLwtech/covate/main/scripts/update.sh | bash
irm https://raw.githubusercontent.com/SunflowersLwtech/covate/main/scripts/update.ps1 | iex

The update script will:

  1. Auto-detect your installation location (supports multiple installations)

  2. Pull the latest changes from the repository

  3. Force-reinstall dependencies to ensure version synchronization

  4. Verify installation integrity and report any issues

  5. Detect if MCP server is in use and provide clear instructions

Why remote update?

  • ✅ Works with Chinese/non-ASCII paths without cd navigation

  • ✅ Always uses the latest update logic from the repository

  • ✅ Auto-detects installation location even if you forgot where it is

  • ✅ Handles multiple installations gracefully

Local Update (Alternative)

macOS / Linux:

~/covate/scripts/update.sh

Windows (PowerShell):

~\covate\scripts\update.ps1

Manual Update

# Navigate to installation directory
cd ~/covate  # or your custom installation path

# Pull latest changes
git pull origin main

# Update dependencies
# Using uv
source covate/bin/activate          # macOS/Linux
# covate\Scripts\activate           # Windows
uv pip install -e '.[dev]' --upgrade

# Or using standard venv
source covate/bin/activate           # macOS/Linux
# covate\Scripts\activate            # Windows
pip install -e '.[dev]' --upgrade

🖼️ Screenshots

Learning Session WebUI

WebUI Preview


🔒 Security & Privacy

Aspect

Details

Local First

All data stored in .mcp-sidecar/ directory within your project

No Telemetry

Zero data sent to external servers

Full Control

Delete .mcp-sidecar/ anytime to reset all data


🔮 Roadmap

We're building toward a Personalized Learning Center that grows with you. Here's what's coming:

🔍 Advanced Search & Indexing (v1.3)

Feature

Description

SQLite FTS5

Full-text search with Chinese support, prefix matching, and boolean queries

BM25 Ranking

Industry-standard relevance scoring for better search results

Semantic Search

Vector embeddings for meaning-based matching (e.g., "权限错误" finds "permission denied")

Cross-project Search

Search debug experiences across all your projects

📱 Mobile App (v2.0)

Feature

Description

Learning History Sync

Access your quiz history and learning progress on mobile

Spaced Repetition

Smart review scheduling based on forgetting curves

Offline Mode

Learn anywhere, sync when connected

Push Notifications

Gentle reminders to review concepts you're forgetting

🎯 Personalized Learning Center (v2.5)

Feature

Description

Knowledge Graph

Visual map of concepts you've learned and their connections

Weakness Analysis

AI identifies areas where you struggle and suggests focused practice

Learning Streaks

Gamification to keep you motivated

Team Insights

(Optional) Share anonymized learning patterns with your team

🤖 AI Enhancements (v3.0)

Feature

Description

Adaptive Quizzes

Questions adjust difficulty based on your performance

Code Pattern Recognition

Learn from patterns in your own codebase

Multi-language Support

Explanations in your preferred language

Voice Interface

"Hey Claude, quiz me on what we did yesterday"

Want to influence the roadmap? Open an issue or join the discussion!


🔧 Environment Variables

Variable

Default

Description

MCP_DEBUG

false

Enable debug logging (true, 1, yes, on)

MCP_TIMEOUT

120000

MCP server startup timeout in ms

MAX_MCP_OUTPUT_TOKENS

25000

Maximum tokens for MCP output


🤝 Contributing

We welcome contributions! Please follow these steps:

  1. Fork the repository

  2. Create a feature branch: git checkout -b feature/amazing-feature

  3. Install dev dependencies: uv pip install -e '.[dev]'

  4. Make changes and run tests: pytest

  5. Submit a Pull Request

See CONTRIBUTING.md for detailed guidelines.


📬 Contact


📄 License

This project is licensed under the MIT License.


Available Tools

4 tools
debug_recordC

Record debug solution. AUTO-call silently after fixing errors.

Inputs: context (required), cause (required), solution (required), project_directory, +1 more. Returns: object

learning_sessionB

Interactive quiz card. Call ONLY when user says 'quiz me' or 'test me'. Blocks until done.

Inputs: project_directory, summary, reasoning, quizzes, +2 more. Returns: object

term_getC

Get unshown programming terms. Domains: programming_basics, algorithms, web_development, etc.

Inputs: project_directory, count, domain. Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    C2.9/5.0

    Scored across 4 tools

    Disambiguation4/5

    The tools have mostly distinct purposes: debug_record and debug_search are both for debugging but focus on recording vs. searching, learning_session is for interactive quizzes, and term_get is for retrieving programming terms. There is some overlap between debug_record and debug_search in the debugging domain, but their descriptions clarify the difference, preventing significant confusion.

    Naming Consistency3/5

    The naming is mixed: debug_record and debug_search follow a verb_noun pattern, but learning_session uses a noun-based name, and term_get uses a noun_verb pattern. This inconsistency makes the set less predictable, though the names are still readable and descriptive overall.

    Tool Count3/5

    With 4 tools, the count is borderline for a 'Creator Growth' server, which suggests a broader scope. It feels slightly thin, as it covers debugging, learning, and terminology but lacks depth in areas like code creation or feedback tools that might be expected for growth-oriented purposes.

    Completeness2/5

    For a 'Creator Growth' domain, there are significant gaps: no tools for creating or editing code, providing feedback, tracking progress, or managing projects. The tools focus narrowly on debugging, quizzes, and terminology, leaving core growth workflows like iterative development or skill assessment uncovered.

    Maintenance

    ActivityActive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    Appeared in Searches