Use cases

Most AI tools forget everything the moment a chat ends. Context you've already explained — your stack, your preferences, decisions you made last week — has to be re-explained every session, and it never carries across to a different tool. With Membase, your agents share one persistent memory:

• Persistent project context — your agent remembers your stack, conventions, and past decisions instead of asking again every session.
• Cross-tool continuity — capture something in Claude Code and recall it in Cursor or Codex, because they share one memory.
• Long-running work — pick up a multi-day task with the relevant history pulled back into context automatically.
• Personal knowledge — save facts, preferences, and notes once and let any connected agent retrieve them on demand.

Installation

One command, every detected client (Claude Code, Cursor, Codex):

npx plugins add aristoapp/membase-plugin-mcp

Or Claude Code alone, via its plugin marketplace:

claude plugin marketplace add aristoapp/membase-plugin-mcp
claude plugin install membase@membase-plugins

Per-client manual setup below. Every client connects to the same hosted MCP server: https://mcp.membase.so/mcp.

Claude Code

claude mcp add --transport http membase https://mcp.membase.so/mcp

Want auto-capture and cross-session handoff too? Install the Membase plugin (a bundled MCP server plus hooks) instead — the repo root is the plugin package; see the install commands above or clients/claude.

Full guide: docs/install/claude.md

Cursor

Click to install:

Or add it by hand to .cursor/mcp.json (this project) or ~/.cursor/mcp.json (all projects):

{
  "mcpServers": {
    "membase": {
      "url": "https://mcp.membase.so/mcp"
    }
  }
}

Full guide: docs/install/cursor.md

Codex CLI

codex mcp add membase --url https://mcp.membase.so/mcp
codex mcp login membase

Or add it to ~/.codex/config.toml (global) or .codex/config.toml (this project):

[mcp_servers.membase]
url = "https://mcp.membase.so/mcp"

Full guide: docs/install/codex.md

VS Code

Click to install:

Or add it to your MCP config (.vscode/mcp.json):

{
  "servers": {
    "membase": {
      "type": "http",
      "url": "https://mcp.membase.so/mcp"
    }
  }
}

Hermes Agent

Add the server under mcp_servers in ~/.hermes/config.yaml:

mcp_servers:
  membase:
    url: "https://mcp.membase.so/mcp"

Prefer a native integration? Hermes also ships a Python provider package — see clients/hermes.

Full guide: docs/install/hermes.md

OpenClaw

Remote MCP (simplest). Add Membase to your OpenClaw MCP config:

{
  "mcpServers": {
    "membase": {
      "url": "https://mcp.membase.so/mcp"
    }
  }
}

Native plugin (richer — auto-capture and recall). From an OpenClaw checkout:

openclaw plugins install --link ./clients/openclaw
openclaw plugins enable openclaw-membase
openclaw gateway restart

Full guide: docs/install/openclaw.md

Other MCP hosts (ChatGPT, Gemini CLI, OpenCode, Poke, …)

Any MCP-capable host can connect with the generic remote config plus the OAuth prompt on first use:

{
  "mcpServers": {
    "membase": {
      "url": "https://mcp.membase.so/mcp"
    }
  }
}

Tools

Every connector reaches the same hosted MCP tool set:

add_memory · search_memory · add_wiki · search_wiki · update_wiki · delete_wiki · get_current_date

Clients with a native runtime (Claude Code, OpenClaw, Hermes) make memory automatic: just work, and your sessions are captured, recalled, and handed off for you. Details in each client's install guide.

Anything important, save it yourself and pull it back later — plain language is enough, the agent calls the tools for you:

Remember that we deploy from the release branch, never from main.

What did we decide about the auth flow last week?

Save this: the staging database resets every night at 02:00 UTC.

How it works

Connectors talk to a stable Membase Context API.

Client plugin or MCP config      ← per-client adapter (clients/*)
        │
        ▼
Shared connector core            ← packages/core, packages/connector-sdk
        │
        ▼
Membase Context API              ← stable public surface
        │
        ▼
Private Membase memory engine    ← storage, graph, ranking (not in this repo)

Client-specific behavior stays in clients/*; shared behavior lives in packages/*.

For the full design rationale, see docs/architecture.md and the "where does X live" map in MAP.md.

Note: This repository is the connector/integration layer only. The Membase Context API, memory engine, and their supporting services are a separate, private system and are not part of this repo.

Contributing

Contributions are welcome — new connectors, bug fixes, docs, performance optimization. This is a pnpm monorepo (Node.js 20+, pnpm 11+):

pnpm install
pnpm test    # runtime + contract suites
pnpm check   # the same gate CI runs on your PR

Details in CONTRIBUTING.md; please also review our Code of Conduct.

Security

For security issues, see SECURITY.md for how to report a vulnerability, and docs/security.md for the connector secret-handling and redaction model.

License

Released under the MIT License.