--- title: "How to Add Long-Term Memory to Claude" description: "Every documented way to add long term memory to Claude: built-in memory settings, Projects, custom connectors via remote MCP, and memory MCP servers." canonical: https://past.dev/blog/add-long-term-memory-to-claude date: 2026-09-03 category: Research authors: The past.dev team --- # How to Add Long-Term Memory to Claude Claude ships with [long-term memory](/glossary/long-term-memory) built in. As of September 2026 it is enabled by default on Free, Pro, and Max plans, and it maintains an editable memory that carries context across conversations. To add long term memory to Claude beyond that ceiling, Anthropic documents three extension paths: project knowledge, custom connectors to remote MCP servers, and local MCP servers on Claude Desktop. This guide states the native ceiling first, then walks each path in order of officialness, with the official page for every mechanical claim. It ends with a memory server walk-through and a note for people building their own assistant. ## What Claude remembers on its own Claude's memory summarizes your chats and applies that context to future ones. Anthropic's help center states it is on by default for Free, Pro, and Max plans, while Team and Enterprise plans require an owner to enable it ([Use Claude's chat search and memory](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). Memory is scoped per project: each project keeps "its own separate memory space and dedicated project summary," so context learned in one project stays out of the others. The documented controls sit in one place. Everything remembered is listed under Settings > Memory > Topics, where each item can be edited or deleted; "Pause memory" keeps entries without using them, and "Reset memory" deletes them permanently ([same article](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)). Sensitive subjects such as health, politics, and religious beliefs are excluded by default unless you turn on "Include sensitive topics in memory." Incognito chats keep a conversation out of history and memory entirely. There is also a paste-based flow to import memory from another provider, or export Claude's memory as text, on Free, Pro, Max, and Team plans; Anthropic labels imports experimental ([Import and export your memory](https://support.claude.com/en/articles/12123587-import-and-export-your-memory-from-claude)). Our teardown covers the full mechanics: [how Claude's memory works](/blog/claude-memory). That is the ceiling. The write paths into native memory are your conversations and the paste import, and what it holds is a summary rather than a queryable, dated record. The three paths below each extend a different part of that. ## Projects: the curated path Projects are the most official way to give Claude durable context you choose yourself. Every plan can create them, with a cap of five projects on free accounts, and each project carries uploaded knowledge files plus project instructions ([What are projects?](https://support.claude.com/en/articles/9517075-what-are-projects)). Claude uses that knowledge in every chat inside the project. On paid plans, large project knowledge switches to retrieval so document sets beyond the context window stay usable ([RAG for projects](https://support.claude.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects)). Combined with per-project memory, a project behaves like a small workspace: curated documents plus accumulated context, isolated from your other chats. The limit is equally plain: knowledge is manual. You upload files and you update them. Nothing is written back automatically, and freshness is whatever you last uploaded. ## Custom connectors via remote MCP Custom connectors let Claude call tools on a remote server over the Model Context Protocol. Anthropic documents them on every plan as of September 2026: Free accounts can add one, and Pro, Max, Team, and Enterprise accounts can use them across Claude on the web, Cowork, and Claude Desktop ([Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp)). On Pro and Max the documented steps are short: open Customize > Connectors, choose "Add custom connector," enter the remote MCP server URL, and select "Add." On Team and Enterprise plans only Owners can add custom connectors, from organization settings, and members then connect from Customize > Connectors. Connectors are toggled per conversation from the "+" menu. One architectural fact shapes this path: Claude connects to the server from Anthropic's cloud rather than from your device, so the server must be reachable over the public internet ([same article](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp)). Anthropic also attaches direct warnings: "Only connect Claude to servers built and hosted by organizations and applications you trust," review the permissions a server requests during authentication, and expect that "Malicious MCP servers may include hidden instructions that try to make Claude perform unintended actions." ## Local MCP servers on Claude Desktop The desktop app has a second path the web app lacks. Desktop extensions package a local MCP server as a single-click install: Settings > Extensions > "Browse extensions" lists Anthropic-reviewed tools, and "Advanced settings" accepts custom .mcpb files ([Getting started with local MCP servers on Claude Desktop](https://support.claude.com/en/articles/10949351-getting-started-with-local-mcp-servers-on-claude-desktop)). The split between the two paths is documented plainly: remote connectors run in the cloud and work on web, mobile, Cowork, Desktop, and Claude Code, while desktop extensions run on your computer, can reach local resources like the filesystem, and work only in Claude Desktop and Claude Code ([When to use desktop and web connectors](https://support.claude.com/en/articles/11725091-when-to-use-desktop-and-web-connectors)). A memory that must stay on your machine points to this path. ## Add memory to Claude with a memory MCP server A memory MCP server exposes store and retrieve tools, and Claude decides when to call them. That turns the connector plumbing above into working [cross-session memory](/glossary/cross-session-memory): you tell Claude something in one chat, the server persists it, and a later chat can retrieve it. Two named options, described as what they are: - The reference [memory server from the Model Context Protocol project](https://github.com/modelcontextprotocol/servers/tree/main/src/memory) stores a local knowledge graph of entities, relations, and observations in a JSONL file, and runs via npx or Docker. It is a reference implementation, file-backed on your machine, which makes it a natural fit for the local path on Claude Desktop. - past.dev's workspace MCP server is a remote server for the custom connector path. It serves recall over the memory a past.dev workspace already holds, to signed-in members of that workspace. Connected, Claude can query facts your team ingested into past.dev, with dates and sources attached. It does not write to Claude's own memory, and it is only useful if your team routes data into a past.dev workspace in the first place. The walk-through, on the documented flows: 1. Choose the shape. A local server pairs with Claude Desktop through the local MCP flow above. A remote server is added by URL under Customize > Connectors. 2. Enable it in a conversation from the "+" menu. 3. Test the round trip. Ask Claude to remember a fact, open a new chat, and ask for it back. If the server's tools fire, retrieval works independently of the native memory. 4. Apply Anthropic's cautions before connecting anything that holds real data, and disable write-capable tools you do not need. ## If you are building your own assistant Everything above extends Claude the product; if the thing that needs memory is an assistant you are building, memory becomes an [infrastructure choice](/guides/choose-memory-system) of your own. past.dev is a memory API for that layer: POST /api/v1/ingest takes raw text with its original timestamp in a content field, entity resolution links mentions of the same person across sources, and stored facts carry event time, validity windows, and supersession links. POST /api/v1/recall takes a query and returns ranked, dated evidence with sources and one of four statuses: Supported, Conflicted, NoKnownSupport, or UnknownBecauseDegraded. It runs managed or self-hosted in your own Postgres; start with the [quickstart](/docs/memory-api/quickstart) and read [the benchmarks](/benchmarks) for how recall is measured.