Glasser Launches Pay-Per-Call Data Service for AI Agents

Glasser released a pay-per-call data service for AI agents. According to the disclosure, agents can access about 2,000 paid API endpoints from different providers through a single key, with no subscription required. These numbers form the core selling point of the whole thing: data sources that were previously scattered across multiple providers and required separate subscriptions are consolidated under one key.

The nature of the information source needs to be clarified first. The currently visible disclosure comes from a TestingCatalog post on X, and that post is labeled as a Paid partnership. In other words, this is a promotional release, not a first-party official announcement; the specific provider list and coverage of the roughly 2,000 endpoints are also not disclosed.

According to the disclosure, users can type "set up glasser.ai/SKILL.md" in Claude Code, Codex, Cursor, or another supported agent to begin using it. The agent then automatically installs the Glasser CLI and opens a browser window for the user to log in.

After login, the agent handles the catalog itself: finding an endpoint for the task, checking the call cost, executing the call, and reporting the amount spent. The key to this flow is handing the judgment of "which data source to choose and how much this call costs" to the agent, rather than having a person first dig through documentation and price lists.

There is also an open-source skills library on GitHub that supports tasks such as competitive research, SEO audits, and prospect lists, and can be mounted onto existing agent setups. According to the disclosure, this allows workflows that previously required scraping web pages to be changed to calling paid data sources. However, the specific contents, maintenance status, and license of the skills library are not explained.

The discloser says it tested two tasks, costing $0.038 total, deducted from the $1 free credit. These two calls are the only concrete samples for understanding the service's actual form.

The first was an Apollo lookup for cursor.com, which is said to have returned results in less than half a second, including employee count, funding rounds, investors, department breakdown, and 132 detected technologies, costing $0.026. The second was a DataForSEO search for its own domain, which is said to have found that testingcatalog.com ranks for 2,995 keywords on US Google and returned the positions and search volumes of the top 100 keywords, costing $0.012.

These numbers come from the promoter's self-test, and the sample is only two tasks, so they can only serve as a reference for the service's possible form and cannot be regarded as verification of stability or accuracy.

According to the disclosure, most of these data sources are usually locked behind separate monthly plans for SEO, sales, and company research tools. Glasser's prediction is that most teams need only a small number of queries per data source and would rather pay only for those calls.

This judgment is the service's commercial premise, not a verified fact. Whether it holds depends on whether teams' real query frequency is low enough that pay-per-call is more cost-effective than a subscription; the disclosure does not provide any usage distribution or pricing structure to support this. The pricing structure, billing method, and settlement details for pay-per-call are all not explained.

From the source, the service is intended to let workflows that previously required scraping web pages be changed to calling paid data sources, potentially improving the reliability of agent production pipelines (this is an inference, based on the disclosure's statement that it "allows a workflow that used to scrape pages to be changed to calling paid data sources").

But this inference currently lacks verification. Apart from the two self-test tasks, the service's stability and accuracy in real production environments are unknown; whether the agent's automatic endpoint selection and cost control carry risks of mistaken calls or overspending is also not disclosed.

Whether the service has been officially publicly released, its available regions, and its timeline are all unclear. The free credit is $1, enough for only a small number of calls; use requires Claude Code, Codex, Cursor, or another supported agent; the skills library's task types are limited to scenarios such as competitive research, SEO audits, and prospect lists.

More critically, the information structure itself: the disclosure comes only from a post labeled as a paid partnership, with no first-party official announcement source visible, and the test data comes from the promoter's self-test. This means that descriptions such as "about 2,000 endpoints," "single key," and "no subscription" currently remain at the disclosure level, rather than independently verifiable facts.