Skill Function · Getting Started

    Deploy your first skill in minutes.
    Call it from any MCP-compatible agent.

    A Skill Function is a callable AI endpoint hosted in the cloud. You give it a .md file — your instructions, your expertise, your domain knowledge — and it runs as a protected endpoint on InferX. Your agent calls it. The skill thinks. The intelligence is yours.

    01

    Create your skill

    You have two options:

    Option A — Create from scratch

    Write your skill directly in the dashboard. Give it a name, write your instructions in markdown, and select a backing model. The model is bound to the skill — every caller gets the right model automatically.

    CREATE FROM SCRATCH

    Create skill from scratch in InferX console

    Option B — Import an existing .md file

    Already have a SKILL.md or a folder of .md files? Import it directly. The folder must contain at least one .md file. InferX turns it into a callable endpoint immediately.

    IMPORT FROM SKILL FOLDER

    Import from existing skill folder
    Your skill is private by defaultInstructions never leave the platform.
    02

    Select a backing model

    When creating your skill, choose the model that best fits its task:

    Task typeRecommended model
    Simple classification, routing7B–9B
    Data extraction, summarization14B–35B
    Complex reasoning, analysis70B+
    Code review, generationCoder-optimized models
    Right model per taskThis is what makes Skill Function 70–90% cheaper than running everything through a flagship model.
    03

    Find your skill in My Skills

    Once created, your skill appears in the My Skills section. You'll see the skill name, backing model, endpoint URL, and status. From here you can subscribe, delete, or click through to the skill details.

    MY SKILLS PAGE

    InferX My Skills page
    Subscribe to your skillSubscribing registers it in your MCP tool list so your agent can discover and call it automatically.
    04

    Configure InferX MCP

    Add InferX to your agent's MCP configuration. Once configured, your subscribed skills auto-appear as callable tools — no manual wiring needed.

    Claude Code / claude_desktop_config.json

    json
    {
  "mcpServers": {
    "inferx-skills": {
      "url": "https://api.inferx.net/mcp"
    }
  }
}

    Cursor: Add InferX as an MCP server in Cursor settings. Use the same endpoint URL above.

    OpenClaw / other MCP-compatible agents: Follow your agent's MCP config guide. Point it to https://api.inferx.net/mcp

    Auto-discoveryOnce configured, open your agent and your subscribed skills will appear in the tool list automatically.
    05

    Call your skill

    Your agent calls your skill automatically when the task matches. You can also call it directly via API:

    curl
    curl -X POST https://api.inferx.net/skills/{your-skill-name} \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": "your task here"}'
    06

    Build an orchestrator skill

    optional

    An orchestrator skill acts as a router. Instead of answering directly, it reads your prompt and decides which skill to invoke.

    User prompt
    ↓
    Orchestrator Skill ← reads prompt, decides routing
    ↓
    Routes to the right Skill Expert
    ↓
    Expert answers with full domain knowledge

    To build an orchestrator:

    1. Go to Skills → Create Skill → Create orchestrator skill
    2. Click Select Sub-skills and choose all skills to include
    3. Click Generate Metadata, then create the Orchestrator
    4. Select a lightweight model — routing doesn't need a large model
    5. Subscribe to the orchestrator skill

    CREATE ORCHESTRATOR SKILL

    Create orchestrator skill showing sub-skills section

    SELECT SUB-SKILLS

    Select Sub-skills modal

    Example orchestrator instructions

    markdown
    You are a routing orchestrator. Based on the user's prompt, invoke the correct skill:

- For pricing questions → invoke: saas-pricing-expert
- For code review → invoke: code-reviewer
- For legal analysis → invoke: legal-analyst
- For anything else → answer directly

Always route to the most specific skill available.
    Automatic routingYour agent calls the orchestrator. The orchestrator handles everything else.

    Tips

    • Keep skills focused — one skill, one job. Narrow skills reason better than broad ones.
    • Update anytime — edit the .md file, re-deploy, every caller gets the updated version instantly.
    • Your skill is your property — private by default. Nobody else can call or copy it.
    • Mix models — use a 7B for your orchestrator and 70B for your most complex expert skills.

    The model is just the runtime. The skill is the product.

    Ready to deploy your first skill?

    Free to start. No setup required.

    Create a Skill Function →GitHub →