An MCP setup has three checkpoints you can verify separately. The client loads the configuration, the remote server advertises its tools, and one real call returns data you can inspect. This walkthrough connects the same remote server to Claude Code and Cursor, then runs one creator search with a limit of three results.
The verification ran on macOS on 19 November 2026 with Node.js 26.8.1, Claude Code 2.1.316, and Cursor CLI 2026.11.05. The two client checks used temporary MCP configurations. The search ran once over MCP after tool discovery.
Install the connector
This example uses the hiCreator Agent Skill and MCP server. The page has the current setup instructions and the API key entry point. The documented installer requires Node.js 20 or newer, npm, and an API key from a paid Workspace.
Run the installer in the operating environment where Claude Code and Cursor run. Use the terminal's masked prompt to enter the key.
npx --yes --prefer-online hicreator-agent@latest setup --clients claude,cursor
On native Windows, use npx.cmd in place of npx. The installer checks MCP initialization and tool discovery before writing client configuration. In the verification run for this article, the connector package was version 0.4.4. The API key was supplied to the test processes through an environment variable. For everyday setup, follow the installer prompts or your client's supported credential configuration.
Check Claude Code's connection status
From the environment where Claude Code runs, list the configured MCP servers.
claude mcp list
Look for the server named hicreator if you used the installer. An isolated test configuration can use a different name. In this run, the temporary project entry first showed Pending approval. After that specific server was approved in the test settings, claude mcp list reported Connected for the hiCreator endpoint.
A status check can finish successfully even while a server is pending approval. Read the status on the server's line. In an interactive Claude Code session, /mcp shows the configured servers and their state. Before approving a connection, check that it points to the expected service.
Verify Cursor's tool list
Run the Cursor CLI from the workspace that contains the MCP configuration. Use the identifier shown by the first command as the argument to the second.
agent mcp list
agent mcp list-tools hicreator
In the test, Cursor listed the temporary project entry as needing approval. A temporary user-level configuration then returned live tools and showed search_creators_ai with the arguments platform, query, limit, and filters. The temporary configuration was removed afterward.
The server's separate tools/list response advertised 22 tools at the time of the test. For this walkthrough, the important check is that the client can see the search tool and its current arguments. Tool availability can change as the service evolves.
Make one bounded search
The example asks for US-based YouTube creators who publish small-space home organization tutorials. It uses a structured region filter and caps the response at three candidates. Here are the arguments sent to search_creators_ai in the test.
{
"platform": "youtube",
"query": "YouTube creators who publish practical small-space home organization and storage tutorials with step-by-step demonstrations",
"limit": 3,
"filters": {
"regions": ["US"]
}
}
If you run the example from a chat client, ask it to make one search call with those arguments and show the shortlist for review. Discovery checks the connection and advertised operations. The search is the billable Workspace operation, so set the limit before making the call.
Check the returned fields
The single search returned HTTP 200 in about 6.9 seconds on this run. The response contained three items in data.items, along with request and billing identifiers. This is the observed result summarized without creator identities or account identifiers.
| Check | Observed result |
|---|---|
| Candidates returned | 3 |
| Rows with a YouTube profile URL | 3 |
| Rows with a follower count | 3 |
| Creator regions reported | US, US, US |
For a reviewable shortlist, keep the returned profileUrl, followerCount, and region together. Preserve a missing field as unknown. Record your own notes about relevance in a separate column, using the public profile and published content as evidence. The region field in this run described the creator record. For a requirement about the creator's audience, use audience data from an appropriate source.
Frequently asked questions
Does claude mcp list run a creator search?
It reports the configured servers and their connection status. The separate tools/list operation shows which tools the server advertises. A creator search begins when you call the search tool.
Why can the MCP server be listed while its tools are unavailable?
The client may be waiting for approval of that specific server, or the credential may be unavailable in the environment where the client runs. Check the server's status, endpoint, and credential setup. Then repeat tool discovery.
Do I need to run the search in both clients?
One search is enough for this walkthrough. Verify the connection and tool availability in both clients, then use a single bounded lookup to check the response shape and fields.
What should I do if the response contains fewer than three creators?
Record the number actually returned. Review the supplied profiles and identify any missing data. Use the returned count and field coverage when deciding whether another search is needed.
AI was used to draft and revise this article. The connection checks and the single lookup reported above were run against the live service.
Top comments (0)