Visant Labs MCP

Requires a Visant Labs account (visantlabs.com). Without one, none of these tools work: there is no anonymous mode. Generation tools also spend the account's credits.

API: https://api.visantlabs.com MCP: https://api.visantlabs.com/api/mcp Auth: OAuth 2.1 + PKCE (dynamic client registration) or API key (visant_sk_...)

Connect (Claude-first flow)

Step 1: check for existing credentials

cat ~/.visant/credentials.json

If it exists and has apiKey, skip to Step 3.

Step 2: authenticate

Ask the user for their Visant Labs email and password, then run:

npx visantlabs login -e EMAIL -p PASSWORD

This saves ~/.visant/credentials.json with the permanent apiKey (visant_sk_...).

Step 3: configure the MCP (global, any directory)

claude mcp add --transport http \
  --header "Authorization: Bearer <apiKey>" \
  --scope user \
  visant https://api.visantlabs.com/api/mcp

Do not use mcpServers in settings.json or mcp-remote: the server speaks Streamable HTTP directly.

Step 4: verify

claude mcp list
# should show: visant: https://api.visantlabs.com/api/mcp (HTTP) - ✓ Connected

The tools show up with the mcp__visant__ prefix in the next session.

Available tools (~50)

Account & Auth

Brand Guidelines (16 tools)

Branding Machine (5 tools)

Mockup Machine (5 tools)

Creative Studio (8 tools)

Canvas / Whiteboard (8 tools)

Budget (6 tools)

AI Utilities (free)

House rule: a generated prompt is a saved prompt. Every call to ai-generate-image / creative-generate / mockup-generate / video-generate goes to the project's prompts.jsonl ledger: one line per call, with model/params, refs with each one's role, sha256 of the output, status (approved/discarded), cost, and a note on what went wrong for discarded ones. Record it on the spot: it cannot be recovered later. If visant-mockup-creator is installed, the contract is in its references/prompt-ledger.md.

Document & Image Extraction

Moodboard (3 tools)

Community (2 tools)

When a tool goes down

MCP tools break, and the right response is never to insist or to pretend the result came out:

1. Two attempts, no more. Repeated INTERNAL_ERROR is the server, not the payload; a third try only burns time.
2. Use the alternative and say which one. Never swap silently: whoever reads the result needs to know which path it came through.
3. Report the breakage in the work's .md, with the date and the exact message.

Document Extract (dual mode)

Mode 1: image extraction from a URL

{ "url": "https://www.behance.net/gallery/...", "limit": 80 }

Returns image URLs with width, height and alt text (limit max 200). Works with Behance, Pinterest, Dribbble, portfolios, blogs. Handles lazy-loaded images, srcset and background images. Scope: read.

Mode 2: PDF extraction

{ "pdf_base64": "<base64>", "output": "inline", "include_brand_tokens": true }

Also accepts pdf_path for files on the server. output: "disk" saves a .md, "inline" returns text. include_brand_tokens: true extracts colors, typography and strategy (two phases: algorithmic + Gemini).

Expired token / 401

cat ~/.visant/credentials.json   # take .apiKey

Update the Bearer in the MCP config and restart the session. If the file does not exist: npx visantlabs login.

Usage tips

• Brand context in generation: pass brandGuidelineId to mockup-generate, creative-full and creative-generate to inject colors/typography/voice.
• format: "prompt" on brand-guidelines-get returns text optimized for LLM context, not JSON.
• creative-full is the shortcut: it avoids chaining generate + mockup-generate
  • render by hand.
• Credits: account-usage shows the balance. AI utilities (describe-image, improve-prompt) are free.
• Scopes: OAuth tokens enforce read, write, generate; generation tools need generate.