architect-mcp by InboundCPH - MCP Server

The Architect - Google Slides MCP Server

A State-Aware MCP server that acts as an autonomous Presentation Architect. It manages a local "Digital Twin" of presentations in SQLite databases and generates modern, styled charts via QuickChart.io.

Features

Multi-Project Support: Isolated database files per client (e.g., nike.db, tesla.db)
High-End Visuals: Cyberpunk/Modern charts with gradients and rounded corners
Layout Engine: Understands slide geometry and native master layouts
Self-Contained: All config stored in SQLite - portable projects
Smart Updates: Hash-based change detection to skip unchanged data

Quick Start

Local Development

# Install dependencies
uv sync

# Run the server locally
uv run python -m architect.server

The server will start on http://localhost:8000/mcp/

Railway Deployment

Push to GitHub and connect to Railway
Set environment variables in Railway:
- GOOGLE_CREDENTIALS_BASE64: Base64-encoded service account JSON
- PORT: (auto-set by Railway)
Deploy - Railway will use the Procfile automatically

Google Cloud Setup

Create a Service Account in Google Cloud Console
Enable Google Slides API and Google Sheets API
Download the JSON credentials
For Railway: base64 credentials.json and set as GOOGLE_CREDENTIALS_BASE64
Share your presentations with the service account email

MCP Tools

Workspace & Config

Tool	Description
`list_projects`	List all project workspaces
`set_active_project(name)`	Switch/create project
`get_brand_config`	Get colors, fonts, chart settings
`update_brand_config(key, value)`	Update brand settings
`get_project_history(limit)`	View audit log

Google Authentication

Tool	Description
`authenticate_google`	Auth with Google APIs
`get_service_account_email`	Get email for sharing files

Discovery & Binding

Tool	Description
`scan_deck_context(presentation_id)`	Scan for tagged elements
`register_dataset(key, sheet_url, range)`	Link Google Sheet
`link_element_to_dataset(tag_id, dataset_key)`	Bind element to data
`list_datasets`	Show registered datasets

Creation

Tool	Description
`create_slide(pres_id, layout, content)`	Add slide with native layout
`adjust_layout(pres_id, tag, action, value)`	Move/resize elements

Magic Update

Tool	Description
`smart_update_deck(presentation_id)`	Auto-update all linked charts
`generate_chart_preview(type, labels, values)`	Preview chart URL

Tagging Elements

Add Alt-Text to elements in Google Slides using the format:

CHART::REVENUE - Chart element
IMAGE::LOGO - Image element
TEXT::HEADLINE - Text element

The server scans for elements containing :: in their description/title.

Watchman Sync Protocol

The Architect includes a self-healing state management system that automatically detects and recovers from manual user edits.

How It Works

Revision Tracking: Every scan saves the Google Slides revisionId
JIT State Check: Before any update, compares live vs cached revision
Auto-Repair: If drift detected, automatically rescans to sync state
Orphan Cleanup: Removes elements from DB that users manually deleted

What This Prevents

❌ API 404 errors from updating deleted elements
❌ Stale position data from moved elements
❌ Crashes from deleted slides

Flow Diagram

smart_update_deck()
    │
    ├─► HANDSHAKE: Get live revisionId (lightweight API call)
    │
    ├─► DRIFT CHECK: live_revision != cached_revision?
    │       │
    │       ├─► YES: "Human edits detected!"
    │       │        └─► AUTO-REPAIR: scan_deck_context()
    │       │                └─► Remove orphans, update positions
    │       │
    │       └─► NO: "State is clean"
    │
    └─► UPDATE LOOP: Now safe to iterate and push updates

Pitfall Protections Built-In

The Architect handles common failure scenarios automatically:

Pitfall	Problem	Solution
1. Invisible File	Service account can't access your file (404)	Shows exact email to share with
2. Blurry Charts	Chart doesn't match placeholder size	Uses geometry from DB to render exact dimensions
3. URL Too Long	Complex charts exceed URL limits	Switches to POST API for short URLs
4. Grouped Elements	Tags hidden inside Ctrl+G groups	Recursive scanning dives into groups

Getting the Service Account Email

→ authenticate_google()
→ get_service_account_email()
📧 mcp-bot@project-id.iam.gserviceaccount.com

Then share your presentation with this email as Editor.

Database Schema

Each project .db contains 6 tables:

project_settings - Brand palette, fonts, chart defaults
data_registry - Google Sheet connections
slide_state - Tagged elements with geometry/style
theme_cache - Master layout placeholders
audit_log - Change history
presentation_metadata - Watchman revision tracking

Example Workflow

User: Start working on the Nike Q3 report
→ set_active_project('nike_q3')

User: Scan this deck: [presentation URL]
→ scan_deck_context('...')

User: Register the sales data
→ register_dataset('SALES', 'sheet_url', 'Data!A1:E10')

User: Link the revenue chart to sales data
→ link_element_to_dataset('CHART::REVENUE', 'SALES')

User: Update all the charts
→ smart_update_deck('...')

Environment Variables

Variable	Description
`GOOGLE_CREDENTIALS_BASE64`	Base64-encoded service account JSON
`GOOGLE_CREDENTIALS_JSON`	Raw JSON string (alternative)
`PROJECTS_ROOT`	Path to store .db files (default: `./projects`)
`PORT`	HTTP port (default: 8000)
`HOST`	Bind address (default: 0.0.0.0)

License

MIT