Unif API DocsQuickstart
Connect UnifAPI to your agent, then make the same public-data call over HTTP when you productize.
Use UnifAPI in the order most users need it: connect your AI client to the public-data MCP server, run an agent or skill, sign in when live public-data tools need workspace credits, then use HTTP when you want to productize the workflow.
Option A: connect your agent
Add the public-data MCP server — read-only, OAuth, no API key — and add the agent skills with npx skills add. The open-source agents and skills live at github.com/unifapi-agent/agents.
Connect your agent
Use the path for your client. Each one wires the hosted MCP server for live data; add the agent
skills with npx skills add unifapi-agent/agents.
Paste one prompt (Claude Code, Codex, Cursor, OpenClaw, Goose, Hermes — any agent that manages its own MCP servers). Drop this in and it adds the server and runs the OAuth sign-in:
Install the UnifAPI MCP server for me at https://mcp.unifapi.com — a remote
(streamable HTTP) MCP server. It uses OAuth, read-only public data, no API key.Web connector (Claude, ChatGPT, Perplexity, Grok): add a custom connector pointed at
https://mcp.unifapi.com and authorize over OAuth. ChatGPT needs Developer Mode (Plus/Pro/Team/
Enterprise): Settings -> Connectors -> Advanced -> enable Developer Mode -> Create.
Codex wires MCP and skills separately. Add the MCP server in ~/.codex/config.toml:
[mcp_servers.unifapi]
url = "https://mcp.unifapi.com"then run codex mcp login unifapi, plus npx skills add unifapi-agent/agents for skills.
Claude-compatible plugin hosts (Claude Code, Claude Desktop, OpenClaw) can install skills + MCP together in one shot:
/plugin marketplace add unifapi-agent/agents
/plugin install unifapi@unifapiSee Connect your agent for the full reference and advanced manual MCP setup.
Ask for the task result
Start a new chat or thread and describe the artifact you want. For example:
Analyze these Twitter/X KOLs for an AI developer-tool campaign: @vercel, @shadcn, @rauchg.
Use UnifAPI public data and return price ranges, evidence, confidence, and follow-up questions.Authenticate when the agent asks
OAuth opens only when the agent needs to call UnifAPI MCP tools. Usage is billed to your UnifAPI workspace credits, while model execution stays inside the agent product you already use.
Option B: call the HTTP API
Use direct HTTP when you are building product code, tests, or a custom agent runtime.
Create a workspace and API key
Sign in at api.unifapi.com/sign-in, create an API key, and treat it like a password.
export UNIFAPI_KEY="unif_..."Make a public-data call
Every endpoint is HTTP + JSON. One key, one base URL.
curl https://api.unifapi.com/tiktok/videos/7350810998023949599 \
-H "Authorization: Bearer $UNIFAPI_KEY"A successful response uses the UnifAPI public envelope:
{
"request_id": "req_...",
"data": {
"id": "7350810998023949599",
"title": "im so sick of being tired im so tired of being sick",
"video_description": "im so sick of being tired im so tired of being sick",
"duration": 6,
"author": {
"id": "MS4wLjABAAAA...",
"username": "jennmelon",
"display_name": "Jenn Melon"
},
"like_count": 2002496,
"comment_count": 4350,
"share_url": "https://www.tiktok.com/@jennmelon/video/7350810998023949599"
},
"billing": {
"credits_charged": 1,
"records_charged": 1,
"balance_remaining": 99,
"truncated_due_to_balance": false
}
}List endpoints include data, optional pagination, and billing. Errors come back with HTTP 4xx/5xx and an error object — see Errors.