Generate and Install Skills

If you want to unlock the core value of MCP2Skill, this is one of the most important chapters.

The core purpose of MCP2Skill is to convert MCP tools into reusable Skills, so that an AI Agent no longer has to run directly against a full set of MCP tools every time. On-demand loading reduces repeated context and wasted consumption.

Where Skills Are Generated From

The current version supports two sources:

• Generate from a single MCP service
• Generate from a single workspace

If you only want to export the raw capabilities of one service, generating from the MCP service is a better fit. If you have already filtered tools within a workspace, generating from the workspace is usually more appropriate.

Two Generation Modes for Workspace Skills

When generating a Skill from a workspace, the output structure is determined by the workspace's Aggregate Skill Generation toggle.

Non-aggregated Mode: One Skill per MCP Service

By default, Aggregate Skill Generation is turned off. In this case, previewing, exporting, or installing Skills from a workspace generates multiple independent Skills, one for each member MCP service in the workspace.

The characteristics of this mode are:

• Each MCP service maps to one Skill, with the same name as when generated from that MCP service alone
• SKILL.md contains the tool documentation for that service directly, without generating an additional references directory
• The invocation scripts still point to the workspace endpoint, not the single MCP service endpoint
• Tools disabled at the workspace level are not exposed to these Skills

If multiple MCP services contain tools with the same name, MCP2Skill uses the full tool name from the workspace to avoid confusion during invocation.

Aggregated Mode: One Skill for the Entire Workspace

With Aggregate Skill Generation turned on, generating a Skill from a workspace produces a single Skill that represents the whole workspace.

This mode aggregates the workspace's available tools into one Skill, which is better suited to treating a workspace as a complete task toolbox.

Only aggregated mode uses the workspace's own Skill description. Non-aggregated mode reuses each MCP service's own Skill description.

Why Skills Help Save Tokens

The savings here are not an abstract notion of being "more stable" — the way tools are used fundamentally changes:

MCP Is Direct Tool Exposure

When connecting to MCP directly, the AI faces a set of callable tools. This approach is great for compatibility, but it is not necessarily the most context-efficient path.

Skills Are Activated On Demand

After converting MCP tools into a Skill, the AI usually sees only the Skill's name and purpose description at first. Only when a task actually matches does it go on to read the full SKILL.md, scripts, and reference materials.

MCP2Skill's Value Is Not Just Management

The core value of MCP2Skill is not merely gathering tools into one place — it is converting MCP tools into Skills, so the AI loads the full documentation and resources only when needed.

You can understand the Token savings as three points:

• First, convert the MCP tools that were originally exposed directly into a Skill
• Let the AI see a small amount of descriptive information up front, instead of loading the full documentation and resources from the start
• Read the full content only when it is truly needed, reducing the amount of irrelevant context that enters the model

The former is better suited to long-term reuse and cost control; the latter is better suited to compatible integration with multiple clients.

What to Prepare Before Generating

Before installing Skills, first confirm:

• The target MCP service or workspace is fully configured
• If you want to install to an AI Agent, the target directory has been configured in Settings > AI Agents

If you have not configured an AI Agent directory yet, you can start with previewing only, or export to a local directory.

When to Prefer Generating a Skill vs. Using the MCP Gateway

Common Workflow

1. Open an MCP service detail page or a workspace detail page
2. Click Generate Skill
3. Review the preview content
4. Choose to export to a directory, or install to an AI Agent

The preview step is well worth keeping, because it helps you confirm the content matches your expectations before anything is actually written.

If the workspace is in non-aggregated mode, the preview window shows multiple generated results grouped by Skill; you can expand each Skill to inspect its own SKILL.md and script files.

Recommended Workflow

If you want to use MCP2Skill more reliably, the recommended order is:

1. Organize your MCP services first, making sure connections and tools work properly
2. Organize workspaces by task or capability boundary
3. Choose the appropriate source to generate a Skill
4. Install it to the target AI Agent
5. Keep the gateway path available for other clients to reuse as needed

This way, what you end up with is not a temporary configuration but a capability layer you can reuse over the long term.

Export to a Directory

Suitable for:

• Saving a copy of the Skill output first
• Manually inspecting the file structure
• Handing the result over to another environment

When exporting, MCP2Skill lets you choose an output directory and then writes the generated files into it.

Install to an AI Agent

Suitable for:

• You already know exactly where an AI Agent's Skills directory is
• You want the files delivered directly to the target directory, reducing manual copying

Before installing, MCP2Skill first detects the currently configured AI Agent targets. You only need to select which targets to install to.

When Generating from a Workspace Is a Better Fit

If you have already done the following organization in a workspace:

• Kept only a subset of MCP services
• Filtered the tools
• Want to package capabilities for a specific task

Then generating directly from the workspace is usually more appropriate, because it more closely matches the capability boundary you ultimately want to deliver to the AI Agent.

If the workspace contains multiple MCP services with different purposes, we recommend keeping the default non-aggregated mode. If the workspace itself is a tightly coupled task toolbox, then consider turning on aggregated mode.

When Generating from a Single Service Is a Better Fit

If you just want to preserve a service's standalone capability, for example:

• Packaging a retrieval-type service on its own
• Packaging a code-generation-type service on its own
• Testing a single service in isolation

Then generate from that service first.

What to Do After Generating

After generating or installing, we recommend that you immediately:

1. Open the Skills page to confirm the Skill has appeared
2. Check that the Skill names and count match your expectations
3. Preview SKILL.md and the other files
4. Run an invocation test in the actual AI Agent

Related Chapters

• To configure installation targets: see General and MCP Settings
• To view installed Skills: see Manage Installed Skills
• To organize workspaces first: see Workspace Basics
• To learn about the gateway mode as a second path: see Connect External AI Clients