Use LinkedIn jobs data in ChatGPT or Claude

Know what the MCP server does

The Neuton Actors MCP Server gives an AI client five stable tools: search the live catalog, inspect one Actor, list categories, start a bounded run, and poll an existing run. Newly published Neuton Actors can appear in the catalog without adding each scraper as a separate connection.

Connect the remote server

Use the remote Streamable HTTP endpoint shown on the Actor page:

https://neuton--neuton-actors-mcp-server.apify.actor/mcp

1. Open your client's custom MCP or app settings.
2. Create a remote Streamable HTTP connection with the endpoint above.
3. Authenticate through the client's secure Apify flow or store your Apify token in its protected connector settings.
4. Scan the tools and confirm that the five Neuton tools are visible.

For a generic client that accepts environment-variable placeholders:

{
  "name": "neuton-actors",
  "url": "https://neuton--neuton-actors-mcp-server.apify.actor/mcp",
  "transport": "streamable-http",
  "headers": {
    "Authorization": "Bearer ${APIFY_TOKEN}"
  }
}

Never paste a token into a chat prompt, webpage, screenshot, or repository. Use the connector's secret store. See the current OpenAI MCP app guide or Anthropic MCP documentation for client-specific setup.

Search and inspect before spending

Start with a prompt that explicitly prevents a run:

Find the best Neuton Actor for public LinkedIn AI jobs in London.
Inspect its input, output fields, and current price.
Do not run it yet.

Confirm four things in the inspection response: the Actor belongs to neuton, the Store URL is correct, the example input matches your question, and the price is acceptable. Broad search is suited to discovery; Job Details is for known URLs or IDs; specialist intelligence pays for evidence-backed classification; monitors are for repeated checks.

Approve one bounded run

After inspection, ask for one role, one location, and a small result limit. Keep the charge cap explicit:

Run the inspected Actor for AI engineer jobs in London.
Return at most 10 items and set max_total_charge_usd to 0.25.
If the run is still active, poll that run ID instead of starting another.

The gateway defaults to a $1 cap and rejects caps above $25, but a smaller first cap makes mistakes cheaper. A request returning RUNNING is not a failure. Poll its run ID. Do not start a duplicate merely because results are not ready.

Verify billing mode and evidence

The normal Authorization bearer authenticates the MCP connection and does not fund downstream Actor runs. That default uses gateway billing. Only clients that separately send X-Apify-Customer-Token opt into customer-funded mode. Keep either token in the client's secret store, never in a prompt. The response includes billingMode; treat that field and the live Apify Pricing tab as authoritative before increasing volume:

• customer-funded means the connected Apify account funds the selected Actor at its published Store price.
• gateway-billed means the MCP Actor bills returned workflow results at its published gateway rate.
• Catalog search and Actor inspection do not emit paid workflow-result events.
• Failed runs, diagnostics, duplicates, empty outputs, and unsupported classifications must not be treated as paid business data.

Inspect actorName, status, billingMode, datasetId, itemCount, and the source URLs in the returned rows. A successful run with zero rows can be a valid no-match result; a failed run is not evidence that demand is zero.

Troubleshoot without creating extra runs