Fixmind

Bug fixed. You learned nothing. Your agent patches the code, you accept the diff, and the lesson disappears. Fixmind catches it on the way out - a local MCP server that turns every AI-assisted fix into a lesson you can review later.

Fixmind is a local-first CLI and MCP server. It records short learning lessons after meaningful coding fixes, then reuses reviewed lessons as memory when a new task looks familiar. Everything stays on your machine and connects to your AI agents through the Model Context Protocol (MCP).

Features

• Local by Default: Lessons live in ~/.fixmind/learning.db. No account needed, no external API calls.
• Speaks MCP: Works with Claude Code, Cursor, and Codex.
• Spaced Recall: New lessons come back on a schedule with a real question before the answer.
• Real Diffs: Captures the actual bad and good code from your git diff, not just a vague summary.
• Memory Retrieval: Reviewed lessons can be pulled back into context later, so the agent can reuse the rule instead of relearning the same mistake.

Quickstart

Requirements: Node.js 22.13+

Fixmind uses Node's built-in node:sqlite module. If setup fails with ERR_UNKNOWN_BUILTIN_MODULE: No such built-in module: node:sqlite, upgrade Node to 22.13 or newer.

npx fixmind setup

npx fixmind setup initializes local storage and automatically configures detected installations of Cursor, Claude Code, and Codex. Restart your AI clients after setup finishes.

How It Works

1. Fix bugs like normal: Hand a bug to your agent. It patches it.
2. Agent saves the lesson: If the fix involved actual learning, the agent calls the MCP tool to log the problem, root cause, and the bad/good code.
3. Stored locally: The lesson lands safely in your local SQLite database.
4. Browse & review: Fixmind quizzes you on the lesson over time so you actually retain the knowledge.

The Quality Gate

The MCP server doesn't just blindly accept data; it acts as a strict teacher for your AI agent. Before any lesson is stored, Fixmind runs it through a local validation gate:

• No junk: If the agent tries to save a formatting change, a pure refactor, or a UI tweak without a behavior break, the server rejects it.
• Explain the why: If the agent repeats the symptom as the root cause, the server asks for a deeper explanation. For package or runtime failures, it also requires the boundary behind the error: where code runs, what belongs in the client bundle, and the supported interface between systems.
• Transfer, not recall: The server asks the agent to write transfer questions instead of simple recall questions.
• Superseding mistakes: If a saved lesson turns out to be wrong, the corrected lesson can supersede the old one instead of replacing history.

CLI Usage

Manage your learning library from the terminal.

# Setup & MCP
npx fixmind setup             # Configure detected AI clients
npx fixmind setup --scope project # Scope MCP to the current directory
fixmind mcp                   # Start the MCP server manually (agents do this automatically)
fixmind memory hydration       # Show reviewed lessons relevant to a topic

# Browse Lessons
fixmind list                  # View recent lessons
fixmind list --limit 5        # Show the last 5 lessons
fixmind search "hydration"    # Find lessons by keyword or concept
fixmind stats                 # View recurring mistake patterns

# Review & Learn
fixmind review                # Answer recall questions for due lessons

Local Dashboard

Prefer a GUI? Fixmind comes with a beautiful, local-only web dashboard to view your progress, review lessons, and browse your knowledge base. Lesson prose and review questions support Markdown, including inline code and fenced code blocks.

fixmind dashboard

Runs locally on 127.0.0.1. Use --port 8080 to specify a port or --no-open to prevent auto-opening the browser.

Sync (Pro)

Fixmind is fiercely local-first. But if you want your lessons available across multiple machines, you can opt-in to end-to-end encrypted sync. The server only ever sees ciphertext.

fixmind login
fixmind sync push
fixmind sync pull

License: MIT
Privacy: No telemetry. Your data stays on your machine.