io.github.helinakdogan/agent-magnet MCP Server
io.github.helinakdogan/agent-magnet
Behavioral memory layer that learns your preferences and context across Claude, Cursor, and Codex.
What this kind of tool is for
- Hand your agent a structured way to call a model or an inference API.
- Compare outputs from two different providers on the same prompt.
- Wire retrieval, embeddings, or generation into an existing agent workflow.
How to compare servers like this
- Provider coverage:
- Match the tool's supported providers and models to the ones your project uses.
- Cost exposure:
- Model-backed tools spend real API budget; check how usage and limits are configured.
- Maintenance signal:
- The provider landscape shifts weekly; prefer tools updated within the last few months.
- Config surface:
- Prefer tools whose environment-variable or config keys are documented plainly.
This page carries your copy-paste config: pick your MCP client in the install section on this page and use the copy button.
What is the io.github.helinakdogan/agent-magnet MCP server?
Agent Magnet is an MCP server that provides a persistent behavioral memory layer for AI tools. It automatically learns your project decisions, conventions, preferences, and context across Claude, Cursor, and Codex sessions, eliminating the need to re-explain context in each new conversation. Memory is stored locally by default and can be shared across teams with a paid key.
Agent Magnet solves the problem of AI tools forgetting your project context between sessions. It automatically captures and organizes decisions, conventions, failed attempts, goals, and preferences as you work, then makes that memory available across Claude, Cursor, and Codex. You can mark items as done or forget them to keep memory focused and relevant.
How to install io.github.helinakdogan/agent-magnet
Copy-paste configuration for popular MCP clients.
MAGNET_USER_IDrequiredYour identifier used as the memory key across all tools
MAGNET_OPENAI_KEYOpenAI API key for the reflector and classifier models
MAGNET_LOCAL_MODESet to 1 to use local SQLite storage — no Redis or external services needed
MAGNET_REDIS_URLRedis connection URL for persistent shared storage (overrides local mode)
MAGNET_PROJECT_IDNamespace for memory isolation across projects (defaults to 'default')
{
"mcpServers": {
"agent-magnet": {
"command": "uvx",
"args": [
"agent-magnet"
],
"env": {
"MAGNET_USER_ID": "<YOUR_MAGNET_USER_ID>",
"MAGNET_OPENAI_KEY": "<YOUR_MAGNET_OPENAI_KEY>",
"MAGNET_LOCAL_MODE": "<YOUR_MAGNET_LOCAL_MODE>",
"MAGNET_REDIS_URL": "<YOUR_MAGNET_REDIS_URL>",
"MAGNET_PROJECT_ID": "<YOUR_MAGNET_PROJECT_ID>"
}
}
}
}Tools & capabilities
Tools this server exposes to the agent.
Memory storage and retrieval— Automatically learns and stores project decisions, conventions, watch-outs, failed attempts, goals, and preferencesCross-tool memory sharing— Shares one unified memory across Claude, Cursor, and CodexMemory management— Allows forgetting or marking items as done to prevent memory from accumulating unnecessarilyTeam memory— Enables shared memory across team members with a paid Agent Magnet key
Use cases
- Maintain consistent project context across multiple Claude, Cursor, and Codex sessions without re-explaining conventions
- Share architectural decisions and project conventions automatically with team members
- Track failed attempts and learnings to avoid repeating mistakes in future work
- Keep a persistent record of project goals and preferences that inform AI suggestions
- Reduce context window usage by having the AI reference stored memory instead of re-explaining everything
io.github.helinakdogan/agent-magnet MCP server FAQ
Agent Magnet is a behavioral memory layer that automatically learns your project context, decisions, and preferences, then shares that memory across Claude, Cursor, and Codex so you don't have to re-explain things in each session.
Yes, the core local memory functionality is free and MIT licensed. Team memory sharing requires a paid Agent Magnet key from agentmagnet.app.
Run `pip install agent-magnet` then `agent-magnet init`. The tool automatically detects and configures Claude Desktop and Cursor for you.
No, Agent Magnet stores memory locally by default with no Redis, databases, or API keys required. Team memory sharing is optional and requires a paid key.
It learns decisions, conventions, watch-outs, failed attempts, goals, preferences, and actions taken — automatically categorized and organized.
Yes, you can mark items as done or explicitly forget things to prevent memory from just piling up.
README (reference)
Source of truth, from the repository.
Your AI forgets your project every new session. This fixes that.
What it does
- Learns automatically as you work — no "remember this" required
- Shares one memory across Claude, Cursor, and Codex
- Lets you forget or mark things done — memory doesn't just pile up
Quick start
pip install agent-magnet
agent-magnet init
That's the whole install. No Redis, no keys, local by default.
What it remembers
Decisions, conventions, watch-outs, failed attempts, goals, preferences, and actions taken — one memory, sorted automatically.
Team memory
Shared memory for a team — one person's decision becomes everyone's context. Requires a paid Agent Magnet key from agentmagnet.app.
Works everywhere
Claude, Cursor, Codex — same memory, local by default. agent-magnet init detects and configures Claude Code, Claude Desktop, and Cursor for you.
{
"mcpServers": {
"agent-magnet": {
"command": "agent-magnet-mcp",
"env": { "MAGNET_USER_ID": "your_name" }
}
}
}
</details>
MIT licensed. Contribute — and a ⭐ if this saved your context window.
Related MCP servers

TokRepo — AI Asset Registry
TokRepo MCP: tokrepo_discover, tokrepo_verify, tokrepo_install_plan for reusable AI assets.

Bee governed intelligence for chat, code, research, documents, and memory.

Fidelis Memory
Local memory for AI agents: save and retrieve your notes without forgetting the details that matter.

costwright — agent budget analysis
Static worst-case token-budget analysis for LLM-agent workflows + signed budget certificates.
Run commands on a machine you cannot SSH into. Air-gapped, bastion-only, change-controlled.

io.github.heliosarchitect/wems
Real-time natural hazard monitoring: earthquakes, tsunamis, volcanoes, solar weather, air quality
