Connect Airscale MCP to ChatGPT

Connect Airscale to your AI assistant and prospect without leaving the chat. Search for people and companies, enrich contacts, check credits, run AI web research, and export results as downloadable CSV or JSONL files from a single conversation.

Written by Airscale

June 14, 2026

Getting started

Open your AI assistant's connector settings and add a custom MCP connector.

Paste the Airscale MCP server URL:

If your AI client supports OAuth, you will be redirected to sign in with your Airscale account in the browser. Authorize access, close the browser tab after the success page, and return to your AI assistant.

If your AI client does not support OAuth, use the header-based setup with your Airscale API key:

Generic MCP client configuration:

For OAuth-based clients such as Claude web or ChatGPT custom MCP apps, do not paste an API key. Airscale handles authentication in the browser.

Setup walkthroughs

Need a step-by-step guide for your AI assistant?

• Connect Airscale MCP to Claude

• Connect Airscale MCP to ChatGPT

What you can do

Check credits

Ask your assistant to check the credits available in your Airscale workspace.

This is the safest first test after connecting the MCP server. If the assistant returns your balance, the connector is authenticated and ready.

Search for people

Use natural language or structured filters to find people by name, job title, company, location, or keyword.

For large prospecting workflows, start with a count or a narrow search before exporting. People search supports pagination, but the MCP export tools are the right way to retrieve thousands of rows.

Search for companies

Search companies by firmographic, location, event, intent, technology, and website keyword filters. You can also ask the assistant to discover accepted filter values before running a search.

Enrich contacts and profiles

Airscale MCP exposes the core enrichment tools backed by the Airscale public API. You can find professional emails, bulk emails, mobile phones, personal emails, LinkedIn profile URLs, person profiles, company profiles, and reverse lookups from email or phone.

For paid enrichment, review what the assistant is about to run before approving the tool call.

Run AI web research

Use Airsearch to ask a natural-language research question and optionally extract structured fields.

Airsearch is useful when the answer is not a simple database lookup and needs web research.

Export people or companies

Use export tools when you want a file, especially for hundreds or thousands of rows. Exports are asynchronous jobs: the assistant starts the job, polls status, then returns a download link when the file is ready.

Export tools do not stream thousands of rows into the chat. They write the file in Airscale's Cloudflare infrastructure and return compact metadata:

• export_id

• status

• status_message

• progress_message

• poll_after_seconds

• rows_written

• pages_fetched

• total_estimate

• estimated_max_credits

• actual_credits

• download_url when complete

Large exports can take several minutes. If status says queued, follow status_message and poll again after poll_after_seconds.

How credits work

Free tools

These MCP tools do not charge credits:

• airscale_check_credits

• airscale_count_find_people

• airscale_find_companies_filter_values

Use them to validate the connection, estimate audience size, and discover valid filters before spending credits.

Search and enrichment

Search and enrichment through MCP use the same Airscale credit behavior as the public API.

• airscale_find_people costs 0.1 credits per returned lead.

• airscale_find_companies costs 0.1 credits per returned company.

• airscale_airsearch costs 2 credits per call.

• Email, phone, profile, reverse lookup, and other enrichment tools follow their Airscale API pricing.

Your AI assistant may ask for approval before calling tools, depending on the client. Review the proposed action before approving paid requests.

Exports

Exports are paid because the worker pages through the paid search endpoints to generate the file.

Before an export starts, the tool requires explicit confirmation with confirm_credit_spend: true. The response includes:

• estimated_max_credits: maximum possible spend based on max_rows

• actual_credits: final spend based on rows actually written

For example, exporting 10,000 people or companies has an estimated maximum of 1,000 credits at 0.1 credits per row. If only 8,847 rows are available, the completed job reports 884.7 actual credits.

Recommended workflow

1. Check credits.

2. Use free count or filter discovery where available.

3. Run a small search to inspect result quality.

4. Refine filters with the assistant.

5. Start an export only when the target list looks right.

6. Poll export status until completed.

7. Download the CSV or JSONL file and import it into your CRM or outreach workflow.

Security notes

Airscale MCP never accepts API keys as tool arguments. OAuth clients authenticate in the browser, and header-based clients send the API key as an HTTP authorization header.

Normal tool calls forward the authenticated Airscale API key upstream without storing it. Async export jobs store only a short-lived encrypted credential until the job completes or fails, then remove it from job state.

Export status and file lookup are scoped to the user or workspace that created the job. The generated download URL is protected by a short-lived token, so treat it like a temporary secret and do not share it publicly.

Troubleshooting

The connection works but nothing happens

Start with:

If the assistant returns your balance, the connector is connected correctly.

The browser sign-in completed but the AI looks stuck

Some clients do not automatically close the sign-in tab. Close the tab manually after the confirmation page, then return to your AI assistant.

The assistant cannot find Airscale tools

Make sure the Airscale connector is enabled in the current conversation. Some clients require enabling connectors in each new chat.

If Airscale recently added tools, refresh the connector or reconnect it so the client reloads the tool list. The live Airscale MCP server currently exposes 19 tools: 15 core tools plus 4 export tools.

Export status shows queued and 0 rows

This can be normal for large exports. The queue may need a short time to start, and KV-backed status can be eventually consistent. Follow status_message and poll again after poll_after_seconds.

Export is completed but no file is visible

Ask the assistant to call:

with the export_id. Status polling confirms progress; the file tool returns the final download URL.

Not getting enough results

Try broader filters first, then narrow down. Use standard job title, industry, company size, location, and keyword language. For Find Companies, use airscale_find_companies_filter_values to discover valid values for filters such as industry, city, region, topics, and tech stack.