Higgsfield Skills and MCP: choose your setup

Edit a creative project: start with the Production skills bundle and its app-specific instructions. Its public prompts ask for source files and a check of the required local integration before editing.

Generate media or build brand assets with a coding agent: the higgsfield-ai/skills repository publishes Markdown skills. At the reviewed commit it contains eight skills: generate, soul-id, product-photoshoot, brandkit, marketplace-cards, websites, video-explainer and youtube-thumbnail, each with a higgsfield- prefix.

Connect to Higgsfield cloud tools: use the official MCP setup page. Its connector URL is https://mcp.higgsfield.ai/mcp. Follow the instructions for your selected client.

These are related entrypoints with different scopes. The reviewed GitHub tree has no Blender, After Effects, Resolve or TouchDesigner skill folders. We cannot confirm that its installer installs the eleven Production workflows.

A skill supplies instructions for an agent to follow. A connection or executable tool supplies the ability to act. Higgsfield's repository skills use its CLI for cloud operations; a Markdown skill is not itself an MCP server.

The remote Higgsfield MCP endpoint exposes cloud capabilities. Local creative-app work additionally depends on the matching app integration. Connecting that endpoint alone is not evidence that an agent can edit your Blender scene or Resolve timeline.

For the general distinction, see MCP tools and agent skills. For this product, keep the selected workflow's setup instructions alongside the client configuration.

The official bundle lists these workflow targets. This is a source-checked catalog, not a tested compatibility matrix:

For the GitHub installation path, the installation guide requires Node.js for the npx method and describes Higgsfield CLI setup and authentication. The README names Claude Code, Cursor and Codex, plus agents that load Markdown skills. This does not confirm that every client can operate every local creative app.

For a Production workflow, prepare the relevant creative application, any required license, a copy of your project and source assets, and the workflow's local integration. Verify the connection before requesting edits. Exact OS versions, minimum app versions and a universal local-bridge installation command are UNKNOWN in the sources checked here.

Subscription and credits: the official credits policy says MCP/CLI media generation consumes credits regardless of plan. Website Unlimited/free-generation benefits do not transfer to MCP/CLI. A blanket paid-subscription requirement for access to the Production bundle is UNKNOWN from the checked public documentation. Do not treat that as confirmed free access. Agent subscriptions and creative-app licenses are separate prerequisites.

The command below is the official repository's recommended cross-agent install command. It installs from higgsfield-ai/skills; it is not a verified installer for the entire Production bundle. This command is documentation-based and was not executed for this guide.

1. Read the installation guide and choose the intended agent and installation scope.
2. Run the command only after reviewing the source and accepting the setup changes. Follow its CLI and sign-in instructions.
3. Check skill discovery inside your chosen agent. The official account guidance uses higgsfield account status to inspect authentication, plan and credits; interactive sign-in uses higgsfield auth login.
4. If your task is cloud MCP access instead, use the MCP setup page and the client-specific connection flow. Installing Markdown skills and adding a remote connector are separate operations.
5. For a Production app workflow, return to its official page and verify the required local integration. Do not assume the previous steps installed it.

The official Blender workflow asks for source assets and a working local connection. The TouchDesigner workflow targets an editable network; the Resolve Studio workflow calls for editable grading and preservation of the original grade.

Suggested evaluation, not an executed test: use a duplicate, non-sensitive project. Ask the agent to report the selected skill, available integration and intended output before acting. Approve one small edit, then inspect the saved project in the target application. Check that the result remains editable and the original is intact. A chat response or generated preview alone is not proof of a usable project.

Stop if the app cannot be found, the integration is unavailable, or the workflow asks for unreviewed permissions or spending. Resolve that specific prerequisite before retrying.

Skill installed, creative app unavailable: verify that you selected the Production workflow and its integration. The eight-skill repository does not document the same inventory as the eleven-workflow bundle.

Skill absent from your agent: check the installation scope and discovery instructions for that client. Use our Claude skills guide or Codex skills guide for general discovery context.

CLI account unavailable: inspect account status and follow the official interactive authentication instructions. Do not paste credentials into a task prompt.

Generation consumes credits despite Unlimited: this is consistent with the documented MCP/CLI credit policy. Check available credits before authorizing another job.

Output is only a preview: open the saved project in the target app and check editability. We have not verified rendering, export quality or successful project delivery.

Production workflows can modify creative projects; cloud workflows can submit jobs and consume credits. Before using either, review what files the selected tool can read or write and what assets it sends to a service. Use copies and limit access to the task folder where the integration supports it. These are editorial safety recommendations, not claims about built-in sandboxing.

The checked public pages do not establish one universal transport, port, permission model or API-key requirement for all local integrations. Those details remain UNKNOWN here. Inspect the current per-app instructions before enabling a local bridge; do not copy an unverified socket or server configuration from a similarly named project.

Keep secrets out of project files and prompts. Confirm the output location and any upload or paid-generation action before approving it. Our MCP server security checklist provides a broader permission review.