Skip to content

README.md

Latest commit

 

History

History
138 lines (92 loc) · 5.85 KB

README.md

File metadata and controls

138 lines (92 loc) · 5.85 KB
Raw

LongMemory Integrations

LongMemory integrates through native host extension contracts rather than a parallel memory or transcript-import protocol.

n8n community node

Package: n8n-nodes-longmemory

The node supports Recall, Store, Explain, and Stats operations and declares usableAsTool: true for n8n AI Agents. Credentials are stored by n8n and sent as a bearer token.

pnpm --dir integrations/n8n-nodes-longmemory build

For local n8n development, link or install the package directory using n8n's community-node development workflow. Before publishing, use the official n8n-node lint and provenance-enabled release workflow required by n8n.

OpenClaw and Agent Plugins hosts

Bundle: longmemory-agent-plugin

This is an Agent Plugins 1.0 bundle with:

  • plugin.json validated against the official plugin schema
  • mcp.json connecting to LongMemory's Streamable HTTP MCP endpoint
  • skills/longmemory/SKILL.md teaching safe memory use
longmemory serve --mcp-http
openclaw plugins install ./integrations/longmemory-agent-plugin
openclaw plugins inspect longmemory
openclaw gateway restart

Set LONGMEMORY_API_KEY in the OpenClaw gateway environment. Override mcp.servers.longmemory in operator configuration when LongMemory is not at http://127.0.0.1:7331/mcp.

The same standards-based bundle can be consumed by Agent Plugins 1.0 hosts including Cursor and compatible VS Code agent environments. Prefer this shared bundle over host-specific copies when the host implements the standard directly.

Claude Code

Plugin: claude-code-longmemory

The native Claude Code package provides a namespaced LongMemory skill and starts the installed longmemory CLI over stdio. It contains no hooks or executable plugin code.

pnpm integration:claude:validate
claude --plugin-dir ./integrations/claude-code-longmemory

Codex and ChatGPT desktop

Plugin: codex-longmemory

The native .codex-plugin package includes an Agent Skill and bundled stdio MCP server. The repository catalog at ../.agents/plugins/marketplace.json makes it discoverable as a local Codex marketplace:

codex plugin marketplace add .

Install longmemory from the Plugins Directory after restarting the ChatGPT desktop app. A public ChatGPT plugin submission additionally requires registering the hosted LongMemory MCP endpoint and mapping its plugin_asdk_app identifier in .app.json.

Gemini CLI

Extension: gemini-cli-longmemory

The extension adds startup guidance and launches LongMemory over local stdio.

gemini extensions link ./integrations/gemini-cli-longmemory

Restart Gemini CLI after linking or installing the extension.

Cline, Continue, and LibreChat

Templates: mcp-configs

  • Cline uses local stdio with an empty autoApprove list.
  • Continue uses a standalone .continue/mcpServers/ block for agent mode.
  • LibreChat uses authenticated Streamable HTTP and includes the required Docker-host address allowlist.

See the configuration-pack README for exact placement and deployment notes.

Dify

Dify has a native HTTP MCP tool integration. Start LongMemory with --mcp-http, then add http://127.0.0.1:7331/mcp under Integrations > Tools > MCP with bearer authentication. No duplicate Dify plugin runtime is necessary.

Flowise

Use Flowise's Custom MCP tool with the same Streamable HTTP endpoint and bearer header. This exposes LongMemory tools inside Assistants, Chatflows, and Agentflows without embedding another memory engine.

Agent frameworks

Examples: frameworks

Runnable Python examples use each framework's current native MCP client:

  • CrewAI: Agent.mcps with MCPServerStdio
  • AutoGen: McpWorkbench with StdioServerParams
  • LangGraph/LangChain: MultiServerMCPClient tools passed to create_agent
  • OpenAI Agents SDK: MCPServerStdio attached through Agent.mcp_servers
  • PydanticAI: MCPToolset backed by FastMCP StdioTransport

Each example launches longmemory mcp --project current, handles the Windows npm shim, and includes safe memory-use instructions. Install only the chosen framework's requirements:

cd integrations/frameworks/crewai
py -3.11 -m venv .venv
.\.venv\Scripts\python -m pip install -r requirements.txt
.\.venv\Scripts\python main.py

For hosted deployments, switch the framework's native client to Streamable HTTP and connect it to /mcp. Keep tenant, user, project, agent, and framework identity fixed in server configuration; never derive identity from model-generated tool arguments.

Security

  • Plugin artifacts contain no credential values.
  • The n8n package uses n8n's credential type and request helper.
  • The OpenClaw bundle contains no in-process runtime code.
  • Claude Code, Codex, and Gemini CLI packages contain no hooks or executable runtime code.
  • Local editor integrations use stdio and leave tool approval to the host.
  • Framework examples use native MCP lifecycle management and do not embed LongMemory storage.
  • The LibreChat template resolves its bearer token from LONGMEMORY_API_KEY.
  • MCP writes remain governed by LongMemory's project scope, allowlist, read-only mode, and audit log.
  • External host stores are never modified by LongMemory.