Generate on-brand images and video in Claude Code with Lamina

Use Lamina through a hosted MCP connection in Claude Code. Don’t go hunting for a marketplace plugin. Connect and authorize Lamina, have Claude pull the relevant brand rules before it proposes creative, approve the planned cost, then check the finished image or video against the product brief before distribution.

That distinction prevents a lot of setup drift. Lamina’s integrations guidance says its MCP server works with Claude Code; Anthropic defines a Claude Code plugin as a separate package that can bundle MCP servers, skills, agents, hooks, language servers, and monitors. For creative operations inside the agent loop, a direct MCP connection is enough.

Get brand context first. Generate second. Lamina’s brand kit goes beyond a logo and hex color: palette, typography, voice, do and don’t rules, reference shots, and product-fidelity constraints can all sit there. Let Claude retrieve that context, then give it a precise campaign request. Don’t make it invent a definition of “on brand.”

Is Lamina a Claude Code plugin?

Lamina is documented as a hosted MCP integration for Claude Code, not as a packaged Claude Code marketplace plugin. That is the answer a developer needs: Claude Code can reach Lamina’s capabilities through MCP without a distributable plugin package.

Anthropic’s plugin reference is clear on this. A plugin can contain an MCP server, while Claude Code can also connect to an MCP server directly. Lamina’s integrations page lists Claude Code as a supported MCP client, and its CLI repository sends Claude Code and Cursor integrations to the hosted agent server with OAuth.

Be exact in internal documentation. Say Claude Code connects to Lamina through MCP. Calling it a “Lamina Claude Code plugin” suggests a package format the available documentation does not establish, then people start looking for plugin-install screens instead of an MCP authorization flow.

How do you connect Claude Code to Lamina through MCP?

Add Lamina’s hosted MCP server in Claude Code using the provider’s published setup command, then complete OAuth authorization. Lamina’s agent guidance uses a plan, cost approval, and execution sequence; its CLI repository specifically recommends the hosted MCP server for Claude Code instead of parsing CLI text output.

Use the current Lamina Agents or CLI instructions. Don’t copy an endpoint from an old team wiki. Lamina’s agent page and CLI repository show slightly different hosted endpoint versions, so follow the active provider instruction. After authorization succeeds, Claude Code can invoke Lamina rather than passing creative production off manually.

Scope the connection to the workspace with the right brand kit and approved product material. Chat mention is not the test. Claude should be able to retrieve the intended brand context, propose a constrained asset, show the cost for approval, and return a trackable job.

What should Claude retrieve before it generates an asset?

Claude should retrieve the applicable Lamina brand kit before generating anything. The lookup needs to establish the visual system and product constraints behind an acceptable output, not just return a brand name or folder label.

For a catalog image, bind the correct product reference, category, colorway, materials, silhouette, crop, and intended marketplace placement. For a vertical reel, add the opening frame, duration expectation, motion direction, text-safe areas, voice, and any claim needing legal or merchandising approval. Lamina lists product shoots, vertical reels, ad variants, virtual try-on, campaign banners, and brand films as brand-aware output types. Pick the asset class before asking for variation.

Reference shots matter because they turn taste into a target. Pair each one with explicit do and don’t rules: preserve the garment’s shape, avoid extra hardware, keep product color truthful, do not add unsupported copy, and do not let an atmospheric background obscure the SKU. The generator can handle complex styling and believable material detail. The brief decides whether it makes the right product.

How should you prompt Claude Code for on-brand image and video work?

Give Claude Code a structured creative ticket: business decision, product truth, channel format, and the brand rules it must follow. Tell Claude to run brand lookup first, summarize the constraints it will apply, propose the Lamina run, and wait for approval before execution.

A practical request can read: “Retrieve the brand kit for the summer outerwear campaign. Create a vertical product reel concept for the approved jacket reference. Preserve its olive color, quilted construction, collar shape, zipper placement, and logo treatment. Use the approved editorial voice, leave room for the campaign line, exclude pricing claims, and return a plan with cost before creating.” Specificity carries the load. Claude turns a merchandiser’s request into a production plan you can review.

Separate immutable facts from variables for image variants. Immutable facts include the SKU, colorway, construction, approved logo, and claims. Variables can cover scene, styling, crop, model context, motion concept, or call-to-action placement. That gives the team campaign range without turning a product test into a product redesign.

Why should cost approval stay inside the agent workflow?

Keep cost approval inside Claude Code because it forces a decision before Lamina executes a metered creative run. Lamina publishes the sequence as plan, approve cost, execute. Keep that order, even for a routine-looking prompt.

Approval matters most on broad variant requests. “Create options” might mean one banner treatment to one stakeholder and a large matrix of formats, scenes, and copy directions to another. Have Claude state the asset type, intended variations, brand context, and expected deliverable before the approver releases the job.

Lamina telemetry reports a median asset generation time of about four minutes. Useful for planning a working session. It is generation time, though, not published-asset time; human review, revision decisions, legal checks, and channel-specific trafficking come after. Keep the approval gate to protect creative spend and reviewer attention.

Rashad Jungbluth, Geschäftsführer at Jungbluth.AI, puts the shift plainly: the agent can call specialized tools in sequence, while the human sets the brief, approves the spend, and judges the result.

Which Lamina tasks work well from Claude Code?

Claude Code works well as the front end for Lamina tasks that need repeatable brand context and explicit production constraints. Lamina says its MCP server exposes creation, tracking, evaluation, distribution, and brand lookup, letting an agent carry a request from brief through review instead of stopping at a prompt draft.

Use product shoots for PDP and catalog needs; ad variants for structured creative testing; vertical reels for social placements; virtual try-on for modeled presentation; campaign banners for channel-specific layouts; and brand films for broader narrative work. Those are Lamina’s stated output categories. They do not carry identical review requirements.

Model routing matters. The creative job changes from request to request. Lamina says it routes across image, video, and try-on models, so tell Claude what to optimize: product fidelity, a particular motion treatment, an on-model visualization, or a specific campaign format. “Generate a great asset” is useless as an operational instruction.

What must you review before publishing a generated asset?

Review product truth, brand compliance, and channel readiness before publishing any Lamina output. A human art director or merchandiser should sign off on the final result, especially for a hero placement, paid ad, launch asset, or product detail page where a wrong color, construction detail, or claim creates downstream risk.

Check the approved source for silhouette, material behavior, color, logos, closures, trims, scale, and visible inclusions. Then check palette, type treatment, voice, prohibited motifs, reference-shot alignment, and any required legal line. For video, inspect the opening frame, frame-to-frame continuity, text-safe areas, crop behavior, and whether motion distorts the SKU.

Use Lamina’s evaluation and tracking capabilities in that handoff. They do not replace judgment. The standard is simple: a reviewer should identify the product, recognize the brand, and approve the channel treatment without explaining away a visible inconsistency.

What are the limits of this Claude Code workflow?

This workflow needs a precise brief and usable brand context in the workspace. A weak reference image, vague campaign intent, missing product constraints, or ambiguous approvals simply introduces ambiguity earlier in the process.

Generated creative still requires art direction. Brand-critical hero moments need closer inspection, and the creative team should decide which details are immutable before an agent starts making variants. The gain is faster production of on-brand directions from a governed brief, with approval and evaluation kept close to the request.

Keep endpoint configuration current as Lamina updates its hosted MCP guidance. Its agent and CLI pages show different endpoint-version examples, so standardize on the active documented instruction, record who owns the connection, and recheck authorization when a workspace or access policy changes.

FAQ: Claude Code and Lamina MCP

Can Claude Code generate both images and video with Lamina?

Yes. Lamina describes image, video, and try-on model routing through its MCP server, and lists product shoots, vertical reels, ad variants, virtual try-on, campaign banners, and brand films as supported brand-aware output categories. Choose the output type and set review criteria in the brief before Claude proposes the run.

Do you need a Claude Code marketplace plugin to use Lamina?

No. Lamina’s published guidance supports Claude Code through a hosted MCP connection, while Anthropic distinguishes an MCP connection from a packaged Claude Code plugin. Configure the MCP integration and complete OAuth authorization.

Can Claude Code publish Lamina assets automatically?

Lamina says its MCP server includes distribution capability, though automatic publication should follow human approval. Use tracking and evaluation to confirm product fidelity, brand rules, claims, and channel formatting before routing an asset to a live destination.

What is the first prompt to send after connecting Lamina?

Ask Claude to retrieve the relevant Lamina brand kit and summarize the palette, typography, voice, reference direction, do and don’t rules, and product-fidelity constraints it will apply. Then provide the SKU, channel, asset type, and campaign objective for a planned run.