# GoodMem Claude Code Plugin

A Claude Code plugin for [GoodMem](https://docs.goodmem.ai) — memory infrastructure for AI agents.

With this plugin, you can operate GoodMem's memory infrastructure in plain English — create and manage embedders, rerankers, and LLMs, ingest memories from files, base64 strings, and plain text, and build knowledge-powered agents for RAG and Deep Research. Also includes Python and Java SDK references for writing GoodMem code.

## Use cases

- **Deep Research agent** — Create an embedder and space, ingest research papers and documents, then ask complex questions that retrieve and synthesize information across your entire knowledge base.
- **RAG pipeline** — Configure embedders, rerankers, and LLMs, store source materials as memories, and retrieve relevant context for answering questions.
- **Self-improving support agent** — Ingest historical issue tickets and resolutions, search for solutions matching a new problem description, and iteratively store new discoveries and fixes as memories so future troubleshooting gets smarter over time.

## Installation

**From the community marketplace** (preferred):

```
/plugin marketplace add anthropics/claude-plugins-community
/plugin install goodmem@claude-community
```

**From this repo** (fallback — use only if the community marketplace is unavailable):

```
/plugin marketplace add PAIR-Systems-Inc/goodmem-claude-code-plugin
/plugin install goodmem@goodmem-plugins
```

## Configuration

### What you need to configure

**GoodMem credentials** (required):

- `GOODMEM_BASE_URL` — GoodMem server URL (e.g., `https://your-server.example.com`)
- `GOODMEM_API_KEY` — GoodMem API key (starts with `gm_`)

**TLS / self-signed certificates** (optional):

If the server uses a private or self-signed CA, set `NODE_EXTRA_CA_CERTS` to the absolute path of its PEM certificate file before launching Claude Code. Restart the MCP connection after changing this setting. An expired or invalid server certificate must be fixed on the server.

### How to configure

**1. In English from chat** (GoodMem credentials only; TLS variables cannot be set this way)

Just tell Claude your server details — it will call `goodmem_configure` automatically:

```
> "Configure GoodMem with base URL https://my-server.com and API key gm_abc123"
```

Credentials persist for the session. You can reconfigure anytime to switch servers. 


**2. Shell export** (before launching Claude Code; works for all configurations)

```bash
export GOODMEM_BASE_URL="https://your-server.example.com"
export GOODMEM_API_KEY="gm_..."
export NODE_EXTRA_CA_CERTS="/path/to/rootCA.pem"  # only if needed for self-signed certs
```

## Reloading after updates

After installing a new version or pulling the latest changes, run `/reload-plugins` inside Claude Code to pick up the updates without restarting. If you are using the plugin from this repo, you can run `git pull` to get the latest changes.

## What's included

| Component | Description |
|-----------|-------------|
| `skills/help/` | Setup guide, available skills overview, example workflows |
| `skills/python/` | Python SDK reference — method signatures, parameters, examples |
| `skills/java/` | Java SDK usage reference — setup, builders, streaming, pagination, examples |
| `skills/mcp/` | MCP workflow guide; tool discovery provides current parameters |
| `.mcp.json` | MCP server with auto-inference from the built-in model registry |

### Skills

- **`goodmem:help`** — Overview of all skills, setup instructions, example workflows
- **`goodmem:python`** — Python SDK reference for writing GoodMem code
- **`goodmem:java`** — Java SDK reference for writing GoodMem code
- **`goodmem:mcp`** — MCP tools reference for direct operations

### MCP tools

The MCP server exposes all GoodMem API operations as tools:

- **goodmem_configure** — set server credentials from chat
- **goodmem_lookup_model** — look up model info from the built-in registry
- **embedders** — create, list, get, update, delete embedding models
- **llms** — create, list, get, update, delete LLM configurations
- **rerankers** — create, list, get, update, delete reranker models
- **spaces** — create, list, get, update, delete memory spaces
- **memories** — create, list, get, update, delete, retrieve, batch operations
- **ocr** — extract text from documents
- **users, apikeys, system, admin** — manage users, API keys, server

### Auto-inference

When creating embedders, LLMs, or rerankers, just provide `model_identifier` and the plugin auto-fills provider, endpoint, dimensions, and other fields from the built-in model registry. You only need to provide `display_name`, `model_identifier`, and credentials — everything else is inferred.

Explicit values always override inferred defaults.

### Credential validation

SaaS providers (OpenAI, Cohere, Voyage, Jina, and OpenAI-compatible endpoints for Anthropic, Google, Mistral) require API credentials. If you create an embedder, LLM, or reranker pointing at a known SaaS hostname without providing credentials, the plugin raises a clear error before the request reaches the server. Pass `credentials` (MCP), `api_key` (Python SDK), or the provider API key overload (Java SDK) to proceed.

## Links

- [GoodMem Documentation](https://docs.goodmem.ai)
- [Python SDK on PyPI](https://pypi.org/project/goodmem/)
- [MCP Server on npm](https://www.npmjs.com/package/@pairsystems/goodmem-mcp)