---

name: integrations-plugins description: "Use Mem0 framework, agent, and editor integrations: Vercel AI SDK provider, MCP/plugin setup for Claude Code, Cursor, Codex, OpenCode, Antigravity, OpenClaw, Pi Agent, lifecycle hooks, slash commands, and framework cookbooks." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

integrations-plugins

Use this sub-skill when a task mentions Mem0, mem0ai, a memory layer, persistent agent memory, Vercel AI SDK memory, @mem0/vercel-ai-provider, createMem0, MCP setup, Claude Code/Cursor/Codex/OpenCode/Antigravity plugins, OpenClaw memory, Pi Agent memory, slash commands, lifecycle hooks, framework agents, or cookbook-style integration work.

Route First

• Use this sub-skill for integration package setup, provider wrapping, framework routes, MCP server config, plugin install choices, lifecycle hooks, slash commands, OpenClaw, Pi Agent, and agent/editor troubleshooting.
• Use Vercel AI Provider for createMem0, addMemories, retrieveMemories, getMemories, searchMemories, AI SDK v6 routes, streaming, tool calls, sources, file prompts, and user scoping.
• Use Editor and Agent Plugins for Claude Code/Cowork, Cursor, Codex, OpenCode, Antigravity, OpenClaw, Pi Agent, MCP-only vs full-plugin installs, hook behavior, slash commands, and duplicate registration checks.
• Use Framework Integrations for LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, Agno, OpenAI Agents SDK, Google ADK, and similar retrieve-generate-store patterns.
• Use Troubleshooting for install/import, environment variables, duplicate tools, unsupported package versions, plugin hook failures, MCP auth, memory scope, and workflow-specific failures.

Sibling Boundaries

• Route direct Python/TypeScript SDK CRUD/search, MemoryClient, Memory, async clients, filters, exports, and feedback to ../sdk-memory/SKILL.md.
• Route OSS provider, vector store, embedder, LLM, reranker, graph-memory, Qdrant/PGVector dimensions, and optional dependency selection to ../provider-configuration/SKILL.md.
• Route terminal-only Mem0 CLI workflows, mem0 init/add/search/list, agent JSON output, and CLI import/export to ../cli-memory/SKILL.md.
• Route self-hosted Mem0 server, Docker Compose, REST deployment, OpenMemory API/UI/MCP hosting, auth, migrations, and operational backups to ../self-hosted-openmemory/SKILL.md.

Fast Decision Tree

1. Vercel/Next.js route or AI SDK model wrapper? Use createMem0 from @mem0/vercel-ai-provider v3.0.0 with AI SDK v6 and preserve stable user_id or equivalent entity scope.
2. Claude/Cursor/Codex/OpenCode/Antigravity request? Decide between full plugin and MCP-only. Full plugin gives hooks/skills; MCP-only gives memory tools only.
3. Duplicate or missing agent tools? Inspect config first with python scripts/validate_mcp_config.py; do not add a second mem0 server blindly.
4. OpenClaw or Pi Agent? Use the plugin-native memory backend/extension and its own command set; do not replace those workflows with generic SDK snippets unless the task asks for custom app code.
5. LangChain/LangGraph/LlamaIndex/CrewAI/Agents framework? Keep the framework’s agent runner intact and add Mem0 as a scoped memory layer around retrieval and storage.

Bundled Scripts

• python scripts/validate_mcp_config.py --help explains read-only MCP/plugin config checks for Claude, Cursor, Codex, OpenCode, and generic JSON/TOML config files.
• node scripts/check_vercel_mem0_usage.mjs --help statically checks TypeScript/JavaScript files for Mem0 Vercel provider imports, createMem0 usage, user scoping, API key patterns, and unsupported AI SDK hints.

Working Rules

• Never print or hardcode real MEM0_API_KEY, provider API keys, MCP bearer tokens, or personal memory contents.
• Prefer environment variables and config templates; redact secret values in diagnostics.
• Treat hook installers, auto-capture scripts, and plugin setup commands as side-effectful. Use the bundled validators before proposing config edits.
• Do not tell future agents to run original repo examples, tests, or plugin scripts. Use the bundled references and scripts here.
• Confirm destructive memory operations, global-scope deletes, hook rewrites, and project-wide plugin config changes before acting.