Memoria
Local-first shared memory that carries decisions and fixes between AI coding tools
Memoria is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.
- Pricing
- Open source
- Platforms
- CLI, macOS, Windows
- Runs
- Locally
- Version
- 0.16.1
- Listing updated
Similar MCP servers
Install
Run in a terminal
claude mcp add memoria -- npx -y @xinzzu/memoriaAdd to claude_desktop_config.json
{
"mcpServers": {
"memoria": {
"command": "npx",
"args": [
"-y",
"@xinzzu/memoria"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"memoria": {
"command": "npx",
"args": [
"-y",
"@xinzzu/memoria"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"memoria": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@xinzzu/memoria"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"memoria": {
"command": "npx",
"args": [
"-y",
"@xinzzu/memoria"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"memoria": {
"command": "npx",
"args": [
"-y",
"@xinzzu/memoria"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
About Memoria
The README says Memoria watches sessions from tools such as Claude Code, Codex, Gemini CLI, Cursor and VS Code, then extracts facts, decisions, open tasks and fixed errors into a single SQLite store on the user's machine. When a session goes idle or a usage limit is reported, a resume engine writes a marked block into AGENTS.md, CLAUDE.md or GEMINI.md so the next tool starts with the same context.
Agents reach the store through an MCP server offered over stdio or a local HTTP bridge at port 48210. Tools include recall, remember, update_memory, forget, file_context and build_context. Extraction works without any model using built-in heuristics, and can be upgraded to a local Ollama model or an OpenAI-compatible endpoint that the user opts into.
It suits developers who switch between several AI coding assistants and want decisions and unfinished work to follow them without re-explaining. Optional end-to-end encrypted sync between a user's own machines is also described.
Tools this server provides (9)
Taken from the project documentation; the running server may differ.
- recall
- Project-scoped memory recall within a token budget.
- remember
- Stores a memory with ADD, UPDATE or NOOP decisions.
- update_memory
- Edits an existing memory.
- forget
- Removes a memory.
- related
- Finds memories related to a given one.
- project_context
- Returns a compact index of project memories.
- get_memory
- Returns full text for selected memory ids.
- file_context
- Lists fixed errors, decisions and open tasks mentioning a file.
- build_context
- Builds one ready-to-paste context block within a character budget.
What it does
- Cross-tool session capture
- Reads session data from nine AI coding tools, including Claude Code, Codex, Cursor and Windsurf, in near real time.
- Automatic resume block
- Injects open tasks, locked decisions and dirty git files into context files when a session goes idle or hits a limit.
- Pre-limit work snapshot
- Saves files touched, commands run and git state when an AI tool reports a usage limit.
- Local MCP memory tools
- Exposes recall, remember, forget and related tools to MCP clients over stdio or HTTP on localhost.
- Works without a model
- Heuristic extraction runs by default, with optional Ollama or OpenAI-compatible engines.
- Encrypted device sync
- Syncs memory between a user's own machines through end-to-end encrypted bundles, excluding embeddings and secrets.
Good for
- Continue a coding task in a different AI tool after hitting a usage limit
- Recall past architectural decisions before editing a project
- Check previous fixes for a file before changing it
- Share project conventions with a team through a committed brain file
Pricing
Open source
MIT licensed, runs locally at no cost.
Memoria FAQ
Does my data leave my computer?
The README says capture, extraction and search run locally in ~/.memoria. Data leaves the machine only if the user explicitly configures the OpenAI-compatible tier.
Do I need Ollama?
No. The README says heuristics work with zero models, and Ollama is an optional upgrade that is detected automatically.
Which MCP tools are included?
The README lists nine tools: recall, remember, update_memory, forget, related, project_context, get_memory, file_context and build_context.
Is Memoria free?
Yes. Memoria is open source, so you can use and self-host it for free.
How do I add Memoria to Claude Code?
Run this command in your terminal: claude mcp add memoria -- npx -y @xinzzu/memoria
What are the best alternatives to Memoria?
The closest alternatives to Memoria are Memware — Cross-Agent Long-Term Memory, Cross-Agent Memory, Total Agent Memory, memry and Agent Memory MCP.
Where this comes from
- Description: written from the project's README and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.