/ Directory / Playground / Memorix
● Community AVIDS2 ⚡ Instant

Memorix

by AVIDS2 · AVIDS2/memorix

One memory store, many coding agents — context follows you across Claude Code, Cursor, Codex, Windsurf so you don't repeat 'we use pnpm and tsx' six times a day.

Memorix is a small MCP server that backs an append-only, structured memory store and exposes it to every MCP-capable coding agent. Add a fact once (e.g. 'auth lives in /apps/api/src/auth, never the deprecated /legacy/auth') and any agent that connects can read it. Designed to slot in without locking you in — local file backend by default, optional cloud sync.

Why use it

Key features

Live Demo

What it looks like in practice

memorix-mcp.replay ▶ ready
0/0

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json  · Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "memorix-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "memorix"
      ]
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json
{
  "mcpServers": {
    "memorix-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "memorix"
      ]
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit
{
  "mcpServers": {
    "memorix-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "memorix"
      ]
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "memorix-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "memorix"
      ]
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "memorix-mcp",
      "command": "npx",
      "args": [
        "-y",
        "memorix"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json
{
  "context_servers": {
    "memorix-mcp": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "memorix"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add memorix-mcp -- npx -y memorix

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use Memorix

Share project conventions across every coding agent without rewrites

👤 Devs juggling multiple AI coding tools ⏱ ~15 min beginner

When to use: You explain the same monorepo layout to Cursor, Claude Code, and Codex on day 1 and forget which one knows what.

Prerequisites
  • Memorix runningnpx memorix — defaults to ~/.memorix.db
Flow
  1. Seed memories
    Save these facts about my project: package manager pnpm, runtime node 22, db postgres 16, auth in /apps/api/src/auth (not /legacy/auth).✓ Copied
    → 5 memories appended with tags
  2. Recall in a different agent
    (in Cursor with same Memorix) — What conventions are set for this project?✓ Copied
    → Returns the same 5 memories
  3. Update one
    Update memory: now using node 24 (was 22). Keep history.✓ Copied
    → Updated; history preserved as a revision

Outcome: Project context persists across agents and across days.

Pitfalls
  • Memories grow stale — Use decay to mark old memories with low confidence; agents weight accordingly

Capture architectural decisions as queryable memories

👤 Tech leads, staff engineers ⏱ ~20 min intermediate

When to use: You make a call ('we don't use Redux, just zustand') and want it to stick.

Flow
  1. Log the decision
    Save decision: 'Use zustand, not Redux. Reason: simpler API, smaller bundle.' Tag: state-management.✓ Copied
    → Decision saved with tag
  2. Verify enforcement
    Whenever Claude is about to import redux in this repo, reject and quote the decision.✓ Copied
    → Subsequent imports are flagged with the saved reason

Outcome: An ADR-light log that agents actually read.

Pitfalls
  • Memorix is per-user; team-wide ADRs need cloud sync or a checked-in mirror — Use --export ./adr.json and commit; teammates --import to sync
Combine with: github

Hand off a debugging session to another agent

👤 Devs who switch between models mid-task ⏱ ~10 min intermediate

When to use: Claude has been debugging for 2h and you want to swap to Codex without re-explaining everything.

Flow
  1. Snapshot
    Snapshot current debugging context: hypothesis, what we ruled out, next step.✓ Copied
    → Snapshot saved with tag debug-session-<id>
  2. Pick up elsewhere
    (in Codex) — Load debug session and continue from where Claude left off.✓ Copied
    → Codex summarizes the snapshot and proposes next step

Outcome: Painless agent swap mid-task.

Pitfalls
  • Snapshot is too vague; new agent restarts from scratch — Memorix's snapshot tool prompts for: hypothesis, evidence, next-step — fill all three

Combinations

Pair with other MCPs for X10 leverage

memorix-mcp + github

Mirror memories to a checked-in adr.json file for team sharing

Export memories tagged 'adr' to /docs/adr.json and commit.✓ Copied
memorix-mcp + filesystem

Backup the local Memorix DB

Copy ~/.memorix.db to /backups/$(date +%F)/.✓ Copied

Tools

What this MCP exposes

ToolInputsWhen to callCost
save content: str, tags: str[] Persist a fact / decision / snapshot free
recall query?, tags?, limit? Bring relevant context into the current task free
update id, content Correct or refine an existing memory free
decay older_than_days Periodic cleanup free

Cost & Limits

What this costs to run

API quota
Local SQLite — bound by disk
Tokens per call
100–500 per recall
Monetary
Free; cloud sync (if used) varies
Tip
Run decay weekly so recall queries return tight, recent context, not 6-month-old debug notes

Security

Permissions, secrets, blast radius

Credential storage: Local SQLite by default; cloud sync needs API key
Data egress: Local-only by default

Troubleshooting

Common errors and fixes

Recall returns nothing relevant

Check tags match — Memorix uses tag filters AND fuzzy text

Verify: Run with no filter to see all memories
DB locked error

Multiple agents writing simultaneously — Memorix uses WAL; ensure version >= 0.4

Import fails

Mismatched schema between exporting and importing version — upgrade both first

Alternatives

Memorix vs others

AlternativeWhen to use it insteadTradeoff
doobidoo/mcp-memory-serviceYou want richer features like knowledge graph and consolidationHeavier; Memorix is intentionally minimal
OpenAI Memory / Anthropic MemoryYou only use one providerLocked to that provider; Memorix is cross-agent

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills