feat(integrations): minimax-mem0 — Mem0 plugin for the MiniMax Marketplace

A declarative MiniMax plugin (MiniMax Code desktop / MiniMax Agent cloud) that
gives an agent persistent long-term memory via Mem0.

- MCP capability points at Mem0's hosted server (https://mcp.mem0.ai/mcp/,
  streamable-http); no secrets in the package.
- Skill (skills/memory/SKILL.md) tells the agent when to recall, save, update,
  and delete memory.
- Manifest, icon (180x180), and structure follow the MiniMax plugin spec;
  validated against the submission checklist and size limits.

Open item for MiniMax review: the MCP endpoint is authenticated (OAuth or
API-key bearer), so confirm whether it ships as a remote MCP (runtime handles
the OAuth handshake) or via the App/Connector path.

integrations/minimax-mem0/.minimax-plugin/plugin.json

@@ -0,0 +1,18 @@
+{
+  "schemaVersion": 1,
+  "name": "mem0",
+  "displayName": "Mem0 Memory",
+  "version": "0.1.0",
+  "description": "Persistent long-term memory for your agent: remember preferences, decisions, and project context across sessions, and recall them on demand.",
+  "author": "Mem0",
+  "icon": "icon.png",
+  "category": "Productivity",
+  "exampleQueries": [
+    "Remember that I prefer TypeScript and pnpm",
+    "What do you remember about my project?",
+    "Update what you know about my deployment setup"
+  ],
+  "apps": [],
+  "mcpServers": ["mem0.mcp.json"],
+  "skills": ["skills/memory/SKILL.md"]
+}

integrations/minimax-mem0/README.md

@@ -0,0 +1,40 @@
+# minimax-mem0
+
+The [Mem0](https://mem0.ai) plugin for the MiniMax Marketplace (MiniMax Code desktop and MiniMax Agent cloud).
+
+It gives a MiniMax agent persistent long-term memory: it remembers the user's preferences, decisions, and project context across sessions and recalls them on demand.
+
+## What's in the package
+
+This is a declarative MiniMax plugin (no build step). It follows the MiniMax plugin spec:
+
+```
+minimax-mem0/
+  .minimax-plugin/
+    plugin.json          # manifest (schemaVersion 1)
+  icon.png               # 180x180 square Mem0 icon
+  mem0.mcp.json          # remote MCP: https://mcp.mem0.ai/mcp (streamable-http)
+  skills/
+    memory/
+      SKILL.md           # when to recall / save / update memory
+```
+
+Two capabilities:
+
+- **MCP** (`mem0.mcp.json`) points at Mem0's hosted MCP server (`https://mcp.mem0.ai/mcp`), which exposes the memory tools (`add_memory`, `search_memories`, `update_memory`, `delete_memory`, and more). No secrets are stored in the package.
+- **Skill** (`skills/memory/SKILL.md`) tells the agent when to search, save, update, and delete memory so recall happens automatically.
+
+## Authentication
+
+Mem0's MCP server is authenticated. It supports OAuth browser sign-in (the client authorizes the user's Mem0 account on first use) or an API-key bearer token for headless clients. No key is committed in this package.
+
+> **Open item for MiniMax review:** because the MCP endpoint requires user authentication, confirm with the MiniMax team whether this is accepted as a remote MCP (the runtime handles the OAuth handshake) or should go through the App/Connector integration path per the plugin guide. See [Prerequisites for App integration].
+
+## Submitting
+
+Submit this directory as the plugin root. For a GitHub submission, use the repo `https://github.com/mem0ai/mem0` with the Plugin subdirectory `integrations/minimax-mem0`. The subdirectory root directly contains `.minimax-plugin/plugin.json`, as required.
+
+## Links
+
+- Mem0: https://mem0.ai
+- Mem0 MCP docs: https://docs.mem0.ai/platform/mem0-mcp

integrations/minimax-mem0/mem0.mcp.json

@@ -0,0 +1,11 @@
+{
+  "schemaVersion": 1,
+  "mcpServers": {
+    "mem0": {
+      "type": "streamable-http",
+      "url": "https://mcp.mem0.ai/mcp/",
+      "description": "Mem0 long-term memory: add, search, update, and delete memories so the agent remembers context across sessions.",
+      "timeout": 30000
+    }
+  }
+}

integrations/minimax-mem0/skills/memory/SKILL.md

@@ -0,0 +1,41 @@
+---
+name: memory
+description: Use Mem0 to persist and recall long-term memory across sessions - the user's preferences, decisions, project context, and durable facts.
+---
+
+You have access to Mem0 long-term memory through the `mem0` MCP server. Use it so the
+user does not have to repeat themselves across sessions.
+
+## When to recall (search memory)
+
+At the start of a task, and whenever the user references something they expect you to
+already know ("my project", "like last time", "the usual"), search memory before asking.
+
+- Call the memory search tool with a short natural-language query describing what you
+  need (e.g. "preferred language and package manager", "deployment process").
+- Use the returned memories as context. Do not announce that you searched; just apply
+  what you found.
+- If nothing relevant comes back, proceed and ask the user only for what is missing.
+
+## When to save (add memory)
+
+Save durable facts as they come up, without being asked:
+
+- Stated preferences ("I use pnpm", "always use TypeScript", "no comments in code").
+- Decisions and conventions ("we deploy on Fridays", "the API base is X").
+- Stable facts about the user, their team, or their project (stack, roles, goals).
+
+Do not save transient or trivial details (a one-off value, a temporary path, secrets,
+or anything the user would not want remembered). Save one clear fact per memory, phrased
+so it makes sense on its own later.
+
+## When to update or delete
+
+If a new statement contradicts something you recalled, update that memory rather than
+adding a duplicate. If the user says to forget something, delete it.
+
+## Boundaries
+
+- Never store secrets, credentials, API keys, or sensitive personal data.
+- Memory is scoped to the connected Mem0 account; keep one user's memories to that user.
+- Prefer recalling silently over interrupting the user to confirm what you already know.