henri-photo

Photographic intelligence via Model Context Protocol. HENRI analyzes, curates, and critiques photographs with museum-curator eye.

Named after Henri Cartier-Bresson. Reads through formal analysis, historical literacy, and curatorial judgment.

Install

Add to your Claude Code MCP settings:

{
  "mcpServers": {
    "henri": {
      "command": "npx",
      "args": ["henri-photo", "--dir", "/path/to/your/henri-data"]
    }
  }
}

Or run directly:

npx henri-photo --dir ~/.henri

Setup

HENRI reads from a data directory containing your analyzed photo archive:

~/.henri/
  manifest-v2.json      # Image manifest with analysis data
  collections/          # Curated thematic collections
  essays/              # Visual essays (critical prose)
  pairs.json           # Cross-body visual pairings
  embeddings/          # CLIP embeddings, color, scene data
  eye/                 # Secondary camera manifest

To build this data, use the HENRI CLI to analyze your photos:

# Clone the repo for full CLI access
git clone https://github.com/sethgoldstein/henri
cd henri && npm install

# Set your data directory and photo source
export HENRI_DIR=~/.henri
export HENRI_PHOTOS=~/Pictures/photos

# Analyze your archive (~$0.03/image via Claude API)
npm run henri -- analyze --batch

# Generate collections and essays
npm run henri -- curate
npm run henri -- collections --propose
npm run henri -- essay -c <collection-id>

MCP Tools

15 tools available to any MCP-compatible client:

| Tool | Description | |------|-------------| | search_archive | Natural language search across all analysis axes | | get_image_analysis | Full 5-axis analysis for a specific image | | list_collections | All curated thematic collections | | get_collection | Collection details with thesis and sequence | | get_daily_signal | Daily selects, cross-body echoes, rising images | | get_tier_breakdown | Archive quality tiers (exhibition through discard) | | get_cross_body_pairs | Same-day echoes across camera bodies | | get_essay | Read a visual essay | | list_essays | All available essays | | search_by_color | Find images by color, harmony, temperature | | get_shooting_patterns | Temporal patterns, golden hour, burst groups | | find_similar | CLIP-based visual similarity search | | get_scene_breakdown | Scene types, indoor/outdoor, objects, materials | | get_people | Face detection, distinct people, co-occurrences | | get_style_clusters | Visual style clusters and near-duplicates |

The 5-Axis Framework

Every image is scored across five axes:

• Formal -- composition, light, spatial organization, color relationships
• Technical -- sharpness, exposure, focus decisions, processing
• Semantic -- subject, narrative, emotional register, cultural context
• Historical -- photographic traditions, references, lineage
• Curatorial -- tier ranking (exhibition/portfolio/strong/study/discard), 0-100 score

Pricing

• Analysis: ~$0.03/image (Claude API cost)
• Essay generation: ~$0.07/essay
• MCP server: free (reads pre-computed data)

Links

• henri.photos -- landing page and case studies
• henri.photos/ask -- chat with HENRI
• henri.photos/hire -- get started

License

MIT