membridge
MemBridge MCP server lets you extract, format, redact, save, and recall portable AI memories locally.
Extract and categorize memories (identity, projects, stack, preferences, rules) from raw text,
conversations.json, or.brain.json.Format memories into ready-to-paste packs for Claude, ChatGPT, Gemini, DeepSeek, Markdown, or JSON, with secrets redacted by default.
Redact API keys, tokens, and emails from any text.
Save durable memory facts to the local
~/.membridge/brain.json, with optional category and redaction.Recall/search persisted memories by keyword and/or category, optionally formatted for a target AI.
Exports memories as Markdown/MEMORY.md packs suitable for coding agents and other tools that use Markdown-based memory files.
MemBridge — Universal AI Memory Teleport
Move your identity, active projects, tech stack, preferences, and rules across ChatGPT, Claude, Gemini, DeepSeek, Grok, and coding agents in 1 click — 100% locally with zero dependencies.
Why MemBridge?
Switching from ChatGPT -> Claude -> Gemini -> DeepSeek shouldn't force you to re-explain who you are, your tech stack, and every active project from scratch.
Feature | One-off Scripts | Vector DB / Cloud Memory | MemBridge v1.1 |
Non-technical Web Studio (zero install) | No | No (needs DB/cloud account) | Yes — membridge.vercel.app |
Direct | Manual unzip | Manual unzip | Built-in browser unzip (0 deps) |
Works across all major AIs | Usually 1 -> 1 | Coding tools only | N-to-N (Claude, ChatGPT, Gemini, DeepSeek, MCP) |
Persistent cross-agent MCP memory | No | Cloud / heavy setup |
|
Optional AI Smart Compress | No | Paid cloud | Local Ollama or BYO Key (direct from browser) |
Auto-redacts API keys & emails | No | No | Built-in secret masker ( |
Dependencies & Telemetry | Heavy | Cloud tracking | 0 npm dependencies · 100% local |
Related MCP server: Memlord
Quick Start (4 Ways to Use)
1. Zero-Install Web Studio (Easiest for Everyone)
Open membridge.vercel.app, double-click index.html, or run locally:
npm start
# Opens at http://127.0.0.1:4321Import: Drop your official ChatGPT/Claude
.zipexport,conversations.json,.brain.json, or paste the built-in 10-Second Quick-Extract Prompt output from any AI chat.Curate & Redact: Toggle, edit, recategorize, auto-mask accidental secrets, or run optional AI Smart Compress via local Ollama.
1-Click Teleport: Export cleanly formatted for Claude XML, ChatGPT Custom Instructions, Gemini Saved Info, DeepSeek/Grok,
MEMORY.md, or.brain.json.
2. Chrome / Edge Browser Extension (In-Tab Teleport)
Download
membridge-chrome-extension.zipfrom Releases (or use theextension/folder).Open
chrome://extensions(oredge://extensions), enable Developer mode, and click Load unpacked.Visit ChatGPT, Claude, Gemini, or DeepSeek -> click Extract Brain -> switch tabs -> click Inject Into Chat.
3. Developer CLI
# Try the built-in demo
node bin/cli.js demo --to claude
# Convert a ChatGPT export into a Cursor / Antigravity MEMORY.md
node bin/cli.js conversations.json --to markdown --out MEMORY.md
# Convert a .brain.json backup for Gemini
node bin/cli.js membridge.brain.json --to gemini4. Persistent MCP Server (Claude Desktop, Cursor, Antigravity, Windsurf)
MemBridge ships a zero-dependency Model Context Protocol server over stdio with 5 tools:
Tool | What it does |
| Save a durable memory fact into |
| Search saved memories by keyword/category and optionally format for any target AI |
| Extract and categorize memories from text, |
| Convert memories into a ready-to-paste pack for |
| Mask API keys, tokens, and emails in any text |
Add to your MCP config:
{
"mcpServers": {
"membridge": {
"command": "node",
"args": ["/absolute/path/to/membridge/bin/mcp.js"]
}
}
}FAQ & Privacy
Does MemBridge upload my chats anywhere?
No. .zip unzipping, parsing, categorizing, and secret redaction all run 100% locally in your browser (or in Node for CLI/MCP). The optional AI Smart Compress only calls the endpoint you explicitly choose (such as localhost:11434 for local Ollama).
Are my API keys safe?
Secret redaction is enabled by default and masks sk-..., ghp_..., AIza..., AWS keys, bearer tokens, and emails before copying, downloading, or saving.
Run Tests
npm testSupport & MemBridge Pro
MemBridge's local Web Studio, CLI, and MCP Server are 100% free & open-source.
Want Encrypted Cloud Sync, Hosted Remote MCP, or Team Memory Packs? Request MemBridge Pro Early Access on membridge.vercel.app/#pro-section or email
hylaw19@gmail.com.Want to support the project? Star the repo ⭐ or reach out for sponsorship / tips.
Credits
Built by Hong Yik (@hyik007) with AI pair-programming via Google Antigravity.
License
MIT — If MemBridge saved you time, a Star ⭐ helps more people discover it!
Available Tools
5 toolsmembridge_extractARead-onlyIdempotent
Extract and categorize portable memories (identity, projects, stack, preferences, rules) from raw text, a ChatGPT/Claude conversations.json export, or a .brain.json. Provide either "text" or "path".
Inputs: path, text.
membridge_formatARead-onlyIdempotent
Convert memories into a ready-to-paste pack for a destination AI. Secrets (API keys, emails) are redacted by default.
Inputs: target (required), path, text, redact.
membridge_recallARead-onlyIdempotent
Retrieve and search persisted user memories from the local portable brain (~/.membridge/brain.json). Use this before answering personalized questions or switching AI tools to recall identity, tech stack, active projects, preferences, or rules. Filter by optional keyword query and/or category, and optionally format the result for a specific target platform (e.g. claude, chatgpt, gemini, markdown, json). Returns JSON with matching memory count and items, or a formatted memory pack when target is provided.
Inputs: query, target, category.
membridge_redactARead-onlyIdempotent
Mask API keys, tokens, and emails in any text.
Inputs: text (required).
membridge_rememberAIdempotent
Save a durable memory fact into your local portable brain file (~/.membridge/brain.json) so it persists across Claude, Cursor, Antigravity, and other MCP clients.
Inputs: text (required), redact, category.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
recall vs remember are clearly read/write opposites, and redact is distinct. However, recall's optional 'target' formatting parameter substantially duplicates membridge_format (both produce a destination-specific memory pack), and extract vs remember have fuzzy boundaries (parse text into memories vs save a fact). An agent could reasonably pick the wrong one for 'give me a Claude-ready pack'.
All five tools use the identical membridge_ prefix with a single lowercase verb (recall, extract, redact, format, remember). No casing or convention mixing.
Five tools is well within the ideal 3-15 range and matches the narrow scope of a portable memory bridge. Each tool maps to a recognizable step: capture, parse, store, retrieve, sanitize, export.
The lifecycle covers create (remember), read/search (recall), import (extract), and export (format/redact). But there is no way to update, delete/forget, or list/clear memories — agents cannot correct a wrong memory or remove stale ones, which is a notable gap for a persistence tool.
Maintenance
Related MCP Connectors
Shared memory MCP server for Claude Code, Codex, Gemini CLI and Cursor. Cited, editable.
Cloud-hosted MCP server for durable AI memory
Persistent memory for AI assistants. Save once; recall from Claude, ChatGPT, or any MCP client.
One memory, every AI. A shared, user-owned markdown memory your AI clients read and write over MCP.
Related MCP Servers
- FlicenseAqualityBmaintenanceA basic implementation of persistent memory using a local knowledge graph. This lets Claude remember information about the user across chats.9340,089 npm92,139-
- AlicenseAqualityBmaintenanceSelf-hosted MCP memory server for personal use and teams1130AGPL 3.0
- AlicenseAqualityDmaintenanceSelf-hosted remote MCP memory server for ChatGPT and AI agents.342MIT
- AlicenseNot gradedqualityDmaintenanceMCP Memory Server for Claude Code that provides persistent context across sessions using semantic search (RAG).Apache 2.0