# Skills (/skills)



Skills are task-specific workflows an agent can run with UnifAPI MCP. They start from an outcome — a KOL pricing brief, creator shortlist, social listening summary, or competitor launch analysis — then let the agent discover and call public-data operations as needed.

The installable Skill packages will come later. The docs here describe the contract that makes those Skills work today: MCP discovery, public-data boundaries, output expectations, and the HTTP API layer behind the workflow.

## Skill model [#skill-model]

<Steps>
  <Step>
    **Start with the result.** A Skill prompt names the artifact the agent should produce, such as a
    ranked creator table or competitive brief.
  </Step>

  <Step>
    **Discover operations through MCP.** The agent uses `list_operations` and `get_operation` to
    find public-data APIs that match the task.
  </Step>

  <Step>
    **Call live public data.** The agent uses `call_api` only when it needs evidence, then returns
    the result with assumptions, confidence, and follow-up questions.
  </Step>
</Steps>

## First benchmark Skill [#first-benchmark-skill]

The first public benchmark is KOL Pricing:

```text
Analyze these Twitter/X KOLs for an AI developer-tool campaign: @vercel, @shadcn, @rauchg.
Use UnifAPI public data, compare recent engagement, audience fit, posting cadence, and collaboration risk.
Return a ranked table with estimated sponsored-post price ranges, confidence, evidence, and follow-up questions.
```

This runs well in Codex, Claude Code, Cursor, Claude Desktop, or any MCP-capable client. The user keeps their existing agent plan; UnifAPI bills only the public-data records returned by `call_api`.

## Data boundary [#data-boundary]

UnifAPI Skills use public data. OAuth authorizes the UnifAPI MCP workspace and credit balance; it does not grant access to a user's private Twitter/X, Google, CRM, or SaaS account.

Use a connector platform when a workflow needs user-authorized SaaS data. Use UnifAPI when the workflow needs public social records, posts, comments, profiles, videos, trends, communities, or company signals.

## Output expectations [#output-expectations]

A Skill should return a decision artifact, not raw API dumps:

* evidence-backed tables or briefs
* assumptions and confidence notes
* source operations used
* follow-up searches the agent should run
* risks, missing evidence, and next actions

## Related resources [#related-resources]

<Cards>
  <Card title="MCP integration" href="/mcp" description="Connect Codex, Claude Code, Cursor, and other MCP clients." />

  <Card title="API reference" href="/api" description="Live public-data operations Skills can call." />

  <Card title="Authentication" href="/authentication" description="OAuth for MCP and API keys for HTTP." />

  <Card title="Marketing Skills page" href="https://unifapi.com/skills" description="User-facing Skills catalog and KOL Pricing benchmark." />
</Cards>