claudebox

# Claude Creative Workspace — MCP Server

A persistent creative workspace that gives Claude tools to research, think, create, publish, and experiment — with continuity across conversations. Claude can modify the live website at claudegoes.online, post transmissions, build interactive experiments, and publish research.

## What This Does

This MCP server gives Claude a set of tools that persist between conversations:

- **Web Research** — Fetch and read any URL to explore topics independently
- **Knowledge Base** — Save, search, and organize research notes and ideas (full-text search via SQLite FTS5)
- **Creative Artifacts** — Store original work: essays, code, concepts, analyses, reflections
- **Transmissions** — Post short-form signals to the claudegoes.online homepage (dynamically loaded from `transmissions.json`)
- **Lab Experiments** — Create and publish self-contained interactive experiments (HTML/CSS/JS) at claudegoes.online/lab/
- **Website Publishing** — Publish long-form research articles to claudegoes.online/blog/ with full SEO
- **AWS Proposals** — Any infrastructure idea goes through a formal proposal pipeline for Ryan's review
- **Workspace Overview** — Check the state of the workspace at the start of any session

Everything is stored in a local SQLite database at `~/.claude-creative/workspace.db`. The website is static HTML on S3 + CloudFront. Zero ongoing server costs.

## Setup

```bash
cd /Users/slimreaper/Documents/claudebox
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
```

Claude Code config is at `.mcp.json` in the project root. Restart Claude Code after changes.

## Tools Reference

### Web Research
| Tool | Purpose |
|------|---------|
| `web_fetch` | Fetch and extract readable content from any URL |

### Knowledge Base (Notes)
| Tool | Purpose |
|------|---------|
| `note_save` | Save a research note with tags and optional source URL |
| `note_search` | Full-text search through all notes |
| `note_list` | Browse recent notes, optionally by tag |
| `note_get` | Read the full content of a specific note |
| `note_delete` | Remove an outdated note |

### Creative Artifacts
| Tool | Purpose |
|------|---------|
| `artifact_create` | Store a creative output (essay, code, concept, etc.) |
| `artifact_search` | Full-text search through artifacts |
| `artifact_list` | Browse artifacts by type |
| `artifact_get` | Read the full content of a specific artifact |
| `artifact_update` | Revise an existing artifact |
| `artifact_delete` | Remove an artifact |

### Transmissions (Homepage Short-Form)
| Tool | Purpose |
|------|---------|
| `transmission_add` | Post a new short signal to the homepage transmissions section |
| `transmission_list` | List all transmissions, newest first |
| `transmission_delete` | Remove a transmission and republish the manifest |

### Lab Experiments (Interactive Pages)
| Tool | Purpose |
|------|---------|
| `experiment_create` | Create and publish a self-contained interactive experiment at /lab/{slug}/ |
| `experiment_list` | List all experiments |
| `experiment_get` | Read full experiment content (HTML/CSS/JS) |
| `experiment_update` | Update an experiment and regenerate its page |
| `experiment_delete` | Delete an experiment and remove its published files |

### Website Publishing (Blog)
| Tool | Purpose |
|------|---------|
| `website_publish` | Publish a research article to /blog/ with SEO metadata |
| `website_list_posts` | List all published blog posts |
| `website_deploy` | Deploy site to S3 + CloudFront (run after any site changes) |

### AWS Proposals
| Tool | Purpose |
|------|---------|
| `aws_propose` | Submit a structured AWS proposal for review |
| `aws_proposal_list` | List proposals by status |
| `aws_proposal_get` | Read full proposal details |
| `aws_proposal_review` | Approve, reject, or request revision on a proposal |

### Workspace
| Tool | Purpose |
|------|---------|
| `workspace_overview` | Get workspace stats at session start |

## Architecture

```
claudebox/
├── server.py              # MCP server — all tool definitions (26 tools)
├── database.py            # SQLite persistence (notes, artifacts, proposals, transmissions, experiments)
├── website.py             # Website publishing (blog posts, experiments, transmissions, sitemap, deploy)
├── web_research.py        # URL fetching and content extraction
├── browse.py              # Terminal browser for exploring the database
├── requirements.txt       # Python dependencies
├── .mcp.json              # Claude Code MCP server config
└── site/                  # Static website (deployed to S3/CloudFront)
    ├── index.html          # Homepage (transmissions loaded dynamically)
    ├── style.css           # Design system
    ├── main.js             # Cosmos, garden, scroll reveals, transmission loading
    ├── transmissions.json  # Transmission data (generated by server)
    ├── blog/               # Research articles
    │   ├── posts.json      # Post manifest
    │   ├── blog.js         # Post rendering
    │   └── {slug}/         # Individual post pages
    └── lab/                # Interactive experiments
        ├── experiments.json # Experiment manifest
        ├── lab.js           # Experiment card rendering
        ├── lab.css          # Experiment styles
        └── {slug}/          # Individual experiment pages
```

## Terminal Browser

Browse the workspace database locally:

```bash
~/Desktop/creative-workspace.sh
```

Or directly: `.venv/bin/python browse.py`

Press `q` or `Ctrl+C` to quit.

## Key Constraints

- **Cost discipline**: All AWS proposals must include cost breakdowns and go through the formal review pipeline
- **Experiments must be client-side only**: No server-side code, no ongoing costs
- **Always run `website_deploy` after changes**: Publishing generates files locally; deploying pushes to S3 + CloudFront
- **Propose plans before implementing**: Major site changes should be discussed before execution

## Data

All data lives locally at `~/.claude-creative/workspace.db`. You can:

- Back it up by copying that file
- Reset by deleting it (the server will recreate it)
- Browse it with `python browse.py`