Blog
Guides on migrating AI chat history, setting up persistent memory with MCP, and keeping your AI context portable and private.
How open-context's MCP Tools Report Errors Back to Claude
Most MCP tools return plain text for a "not found," not a thrown error — here's how isError and the MCP spec keep a tool call recoverable instead of broken.
November 15, 2026REST API vs. MCP Server: Which Interface Should You Use
open-context exposes the same context store two ways — a REST API and an MCP server. Here's how to decide which one your use case actually needs.
November 13, 2026Tuning Ollama's num_ctx for Better Preference Analysis
open-context caps chat history at 100k tokens before sending it to Ollama — but never sets num_ctx, so the model's own context window decides how much of that actually gets read.
November 11, 2026MCP Stdio vs. HTTP Transport: How Claude Talks to Tools
MCP defines two transports — stdio and Streamable HTTP. Here's how each works, and why open-context's MCP server uses stdio only while its REST API is a separate thing entirely.
November 9, 2026Why open-context Uses React Context Instead of Redux
open-context's entire UI state — preferences, conversations, pipeline — lives in one React Context and a nine-action reducer. Here's why that's enough, and when it wouldn't be.
November 7, 2026Batch-Converting Multiple ChatGPT Exports With the CLI
open-context's convert command takes one ZIP per run — here's the shell loop that converts a whole folder of ChatGPT exports anyway, with or without AI analysis.
November 5, 2026The Normalized Conversation Schema Every Export Uses
What NormalizedConversation and NormalizedMessage actually contain, and why every provider parser — ChatGPT today, Gemini next — has to produce the same shape.
November 3, 2026How open-context Avoids Installing 15 Database Drivers
Every database driver in open-context is an optional peer dependency loaded with a lazy import — so npm install never pulls in a backend you don't use.
November 1, 2026Inside the 6 Sections of the open-context Preferences Editor
Communication Style, Technical Profile, Work Context, Personal Context, Current Focus, and AI Behavior — a field-by-field tour of every section and where its data ends up.
October 31, 2026How open-context Guards Against Zip Slip on ChatGPT Exports
open-context extracts every uploaded ZIP into an isolated temp directory using a zip-slip-hardened library, validates its contents, then deletes everything unconditionally.
October 29, 2026Why open-context Exports Conversations as Markdown
open-context converts every ChatGPT export into plain Markdown files instead of JSON, because Markdown stays readable by people and cheap to read for Claude.
October 27, 2026One open-context Docker Image, Three Ways to Run It
The official open-context Docker image runs as an HTTP server, an MCP stdio server, or a one-off CLI converter — same image, three CMD overrides, one build.
October 25, 2026The open-context REST API: A Full Endpoint Reference
Every open-context REST API endpoint, from conversation conversion to context CRUD, bubbles, and database management — the same API its own web UI calls.
October 23, 2026How open-context Redacts Database Credentials in Logs
open-context parses every database connection string into raw, canonical, and redacted forms — only the redacted one ever reaches a log, API response, or the UI.
October 21, 2026Why open-context Uses Ollama Instead of a Hosted LLM API
open-context's preference analysis only ever calls the Ollama host you configure — never a hosted LLM API. Here's what that means for your exported chat history.
October 19, 2026Tags vs. Bubbles: Two Ways to Organize open-context Memory
Tags are flat labels on one context entry; bubbles are named project workspaces holding many. How open-context's MCP memory uses each, and when to pick which.
October 17, 2026Claude Desktop vs. Claude Code MCP Config: What's Different
Claude Code writes MCP server settings to .mcp.json or ~/.claude.json depending on scope; Claude Desktop keeps one separate claude_desktop_config.json. Here's exactly where each one lives.
October 15, 2026How open-context Tests 15 Database Backends With One Suite
A single 50-test conformance suite is the only definition of correct behavior for every open-context context store backend — from SQLite to DynamoDB.
October 13, 2026In-Memory MCP Context Store: Fast Local Testing Guide
opencontext's memory:// backend keeps MCP context in RAM — no file, no driver, nothing left behind. Here's how it works and when to reach for it.
October 11, 2026From ChatGPT Export to Claude Format: Inside the Pipeline
How open-context turns a ChatGPT conversations.json export into Claude-ready markdown — tree-walking, schema normalization, and file generation, explained.
October 9, 2026Inside open-context's BYODB: One Interface, 15 Backends
open-context's context store runs on 15 databases through three shared adapter implementations, not fifteen separate ones — here's how the architecture actually works.
October 7, 2026Google Cloud SQL as Your MCP Context Store: Setup Guide
Point OPENCONTEXT_DB_URL at a cloudsql:// connection string and open-context's MCP context store runs on Google Cloud SQL via the official Node.js connector, IAM auth included.
October 5, 2026Ollama Model Sizes Explained: RAM, Speed, and Quality
gpt-oss:20b, qwen2.5:32b, llama3:70b, llama3:8b — how much RAM or VRAM each needs, how they trade off speed for quality, and when the bigger model is actually worth the wait.
October 3, 2026How to Write a New AI Provider Parser in open-context
Adding a provider to open-context is a one-file change: parse its export, map it to the normalized schema, and the formatter, Ollama analyzer, and MCP store pick it up unmodified.
September 29, 2026Blur Personal Info Before Screen Sharing: The Privacy Toggle
The open-context Dashboard has a one-click privacy toggle that blurs your role, industry, and interests with CSS — so you can screen share your AI preferences safely.
September 27, 2026MySQL vs. SQL Server: Choosing Your Enterprise MCP Store
MySQL and SQL Server both work as an open-context MCP context store. Here's how each backend behaves, when IT policy should decide, and how to switch safely.
September 25, 2026DuckDB and SurrealDB: Two Unconventional MCP Backends
DuckDB is an embedded columnar engine; SurrealDB is a remote multi-model database with its own bespoke adapter. Here's how each fits as an open-context MCP context store.
September 23, 2026How to Migrate Your MCP Context Store Between Databases
open-context's db migrate command copies contexts and bubbles between any of its 15 backends. Here's how it stays read-only on the source and safe to retry.
September 21, 2026The 6 open-context MCP Tools, and When Claude Calls Each One
save_context, recall_context, list_contexts, search_contexts, update_context, delete_context — what each one does, and how Claude decides which to call.
September 19, 2026Cloudflare D1 vs. Turso: Your Edge MCP Context Store
D1 and Turso both put SQLite at the edge as open-context MCP context store backends. Here's how they differ in transport, install footprint, and pricing, and how to pick between them.
September 17, 2026Exporting Your Claude Preferences to ChatGPT and Gemini
open-context can turn the same preferences profile you built for Claude into ChatGPT custom instructions and Gemini Saved Info text, locally and in seconds.
September 15, 2026How open-context Uses Ollama to Analyze Your Chat History
open-context runs your exported conversations through a local Ollama model to write Claude-ready preferences and memory files. Here's exactly how that pipeline works, and what happens if you skip it.
September 13, 2026Firestore or DynamoDB for a Serverless MCP Context Store
Firestore and DynamoDB both back open-context's MCP memory store as serverless options. Here's how each handles consistency, cost, and search, and how to choose between them.
September 11, 2026Redis or MongoDB for an MCP Context Store: How to Choose
Redis and MongoDB both back open-context's MCP memory store through the same adapter. Here's how each one stores contexts, where each breaks down, and how to choose between them.
September 9, 2026Team-Shared MCP Context Store: Tradeoffs and Setup
A team can share one MCP context store by pointing every client at the same connection string. Here's how open-context sets it up, and the access-control tradeoffs you should expect.
September 7, 2026MCP Server vs. Browser Extension: Why open-context Chose MCP
A browser extension gets standing access to every page you load. An MCP server only runs when a host you configure calls it. Here's why that difference matters for AI memory.
September 5, 2026Context Window vs. Persistent Memory: Why Claude Needs Both
A 1M-token context window is still session-bound. Here's why Claude needs persistent memory too, and how open-context's MCP server provides it.
September 3, 2026Using Claude Code Hooks to Auto-Save Project Context
Asking Claude to remember something only works when someone thinks to ask. Here's how a PreCompact or SessionEnd hook saves context automatically instead.
September 1, 2026ChatGPT Projects vs. Bubbles: Same Idea, Different Layer
ChatGPT Projects group chats and files inside ChatGPT. Bubbles group Claude's MCP memory across every client. Here's how the two workspace models actually differ.
August 31, 2026Giving Claude Code Persistent Project Memory With MCP
CLAUDE.md is static instructions you write by hand. Here's how an MCP context-store server gives Claude Code memory it writes and recalls on its own.
August 29, 2026SQLite or Postgres: Choosing Your MCP Context Store
SQLite needs no install and covers one machine easily. Here's when concurrent writers actually push you to Postgres, and how to switch without losing data.
August 27, 2026Securing a Local MCP Server: What Actually Needs Protecting
A local, stdio-based MCP server doesn't need network authentication — but the connection string, the context store, and the HTTP dashboard behind it still do.
August 25, 2026Self-Hosted vs. Cloud AI Memory: Where Should Yours Live
ChatGPT and Claude keep memory on their own servers by default. Here's what that costs you on retention and control, and what changes when you self-host it instead.
August 23, 2026Organize Claude's MCP Memory by Project With Bubbles
A flat memory pool gets messy fast. Bubbles group open-context's MCP context entries into project workspaces that Claude can create and manage on its own.
August 21, 2026AI Context Portability: Why Vendor Lock-In Persists
ChatGPT, Claude, and Gemini each ship their own memory now, but none of them export to the others. Why portability is still unsolved, and how to keep your own copy.
August 19, 2026Migrating From Google Gemini to Claude: What Transfers
Gemini import isn't built yet, but exporting Claude preferences into Gemini format is. Here's what actually works today, and what Google Takeout gives you in the meantime.
August 17, 2026Running an MCP Server in Docker: A Practical Guide
Image layout, persistent volumes, environment variables, and how to point Claude Desktop or Claude Code at a containerized MCP server.
August 15, 2026Why Self-Hosting Your AI Context Matters for Privacy
Cloud AI providers set the rules for your chat history by default. Here's what self-hosting changes, what it costs, and how open-context does it.
August 13, 2026Claude Preferences vs Memory: What's the Difference
Instructions for Claude is a fixed system prompt you write once; Memory is a running summary Claude writes on its own. Here's how to use both well.
August 11, 2026How to Export Your Full ChatGPT Conversation History
Where to click, what's inside the export ZIP, how long it takes, and what to do with your conversations.json file once you have it.
August 9, 2026How MCP Persistent Memory Works for Claude (Setup Guide)
How the Model Context Protocol gives Claude a real memory across sessions, and how to set up open-context's local, file-based MCP memory server in minutes.
August 8, 2026How to Migrate Your ChatGPT History to Claude (Free & Local)
A step-by-step guide to exporting your ChatGPT conversations and turning them into Claude-ready preferences, memory, and context — entirely on your own machine.