Connect UnifAPI MCP to general and coding agents

Connect UnifAPI MCP to general and coding agents
Unif Agents team
Unif Agents team

Updated

Connect UnifAPI MCP to Claude, ChatGPT, Codex, Claude Code, Cursor, VS Code, OpenClaw or Hermes, then run your first public-data Skill.

In short: This is the onboarding article for MCP setup queries: add the hosted UnifAPI server, authenticate with OAuth, and run a Skill from the agent client you already use.

The fastest way to give a compatible general agent, research agent, coding agent, or MCP-capable client access to UnifAPI public data is to connect the hosted MCP server. Add one URL, authenticate when the client asks, and the agent can run Skills without you writing glue code or pasting platform-specific API keys.

The canonical UnifAPI MCP server URL is https://mcp.unifapi.com. Use OAuth where the client supports it. API-key bearer authentication remains a compatibility path, but it should not be the first setup path for most operators.

What you are connecting

MCP gives the agent a tool surface. UnifAPI's hosted MCP server exposes three catalog tools: list_operations for discovery, get_operation for schema and example inspection, and call_api for invoking a concrete public-data operation. The agent does not need a separate tool for every API endpoint.

ThingValue
Server URLhttps://mcp.unifapi.com
Recommended authOAuth through the MCP client
Compatibility authAPI-key bearer header when a client requires static tokens
What gets billedPublic-data records returned by live APIs
What does not get billed by UnifAPIThe model tokens or agent plan you already pay for

Before you start

You need an MCP-capable client, a UnifAPI account when OAuth opens, and enough workspace credits to return public-data records. You can browse Skills before signing in. The sign-in moment should happen when the agent actually needs to call tools.

If you are testing with Claude Desktop or Claude.ai custom connectors, remember that remote connectors are brokered through Anthropic's cloud. Your MCP server must be reachable from the public internet. That is fine for https://mcp.unifapi.com; it is a problem only for private local servers.

Codex

Codex supports remote HTTP MCP servers through the Codex CLI and shared config. Add UnifAPI with:

codex mcp add unifapi --url https://mcp.unifapi.com

Then verify the server is configured:

codex mcp list

Open a new Codex session and ask for a Skill-style task, not just a generic API call. For example: Run a KOL pricing pass for these Twitter/X creators using UnifAPI public data. Return ranges, evidence, confidence, and follow-up questions.

Claude Code

Claude Code's current reference recommends remote HTTP for cloud MCP servers where available. Add UnifAPI from the project where you want Claude Code to use it:

claude mcp add --transport http unifapi https://mcp.unifapi.com

Use user scope if you want the server available across projects:

claude mcp add --transport http --scope user unifapi https://mcp.unifapi.com

Inside Claude Code, run /mcp to confirm connection status and complete OAuth if needed. If the server is project-scoped through .mcp.json, Claude Code may ask for approval before using it.

Claude Desktop and Claude.ai

For remote MCP connectors in Claude Desktop or Claude.ai, use Customize -> Connectors -> + -> Add custom connector. Enter a clear name such as UnifAPI and use the connector URL https://mcp.unifapi.com, then click Connect to authenticate.

Or skip the menus: this link opens the same dialog with the name and URL filled in. Check them, click Add, then Connect.

Do not confuse remote connectors with old local-only Desktop MCP config. Remote connectors are added in the Claude settings UI and can be used across Claude surfaces after they are connected. Local stdio servers configured through desktop config are a separate mechanism.

Cursor

Cursor supports MCP through mcp.json. For a global setup, create or edit ~/.cursor/mcp.json. For a project-only setup, create .cursor/mcp.json in the project.

Use this remote-server shape:

{ "mcpServers": { "unifapi": { "url": "https://mcp.unifapi.com" } } }

Restart Cursor or refresh the MCP tools view if the server does not appear immediately. Cursor's agent uses enabled MCP tools when relevant, and may ask for approval before running a tool depending on your settings.

First Skill to run

Use a task prompt that tells the agent what outcome you want and what evidence to inspect. A good first run is:

Find Twitter/X KOLs discussing AI coding tools and summarize the best outreach angles. Use UnifAPI public data through MCP. Return a ranked shortlist, public evidence, risk notes, and follow-up questions. Keep the workflow bounded to public data.

A second useful test is:

Find the top three TikTok videos about Anthropic this week. Use UnifAPI MCP. Show the public records you used, then summarize the pattern in one paragraph.

The agent should discover operations, inspect the right operation, call it, and synthesize the results. If it tries to guess endpoint names without inspecting the catalog, ask it to call list_operations and get_operation first.

What gets billed

Workspace credits cost $0.001 each. Standard billable records are one credit, and higher-cost operations publish their minimum credits from OpenAPI. A small first top-up is enough to test the shape of a few light workflows. When you are ready for repeated runs, top up a pay-as-you-go balance.

The most important cost habit is to bound the prompt. Ask for top 10 or top 20 records first, then deepen only if the answer is useful. A vague request such as 'research everything about this market' can fan out across many operations. A bounded request such as 'pull 20 Reddit threads and summarize recurring objections' is easier to inspect and budget.

Troubleshooting

SymptomLikely causeWhat to do
Server appears but no tools are usedThe agent did not choose the MCP toolsAsk it to list available UnifAPI operations first
OAuth does not completeBrowser/auth flow did not finishReopen /mcp or the client connector settings and reconnect
Tool call returns unauthorizedWrong auth path or expired tokenReauthenticate with OAuth or check bearer token config
The result is too broadPrompt lacked constraintsAdd platform, time window, record count, and output format
The answer overclaimsPrompt did not set boundariesRequire evidence, confidence, and unknowns

Sources checked

This setup guide was refreshed on June 2, 2026 against the official OpenAI Docs MCP page, Claude Code MCP reference, Anthropic's custom connector help center article, and Cursor's MCP documentation.

MCP setup - keeps the current hosted connector instructions in one place.

Run a KOL pricing agent with MCP - gives a first task after setup.

MCP catalog tool design - explains how agents discover and call operations.