Apify's MCP server hands the agent an Actor schema written for a web form

Two details in the handshake decide whether a hand-rolled client talks to this server at all. The Accept header has to name both `application/json` and `text/event-stream`, or the server rejects the request ^[6]. The session id comes back in a response header instead of the body, so the author's script dumps the headers to a file and reads the id back out of it ^[7]. He used curl on purpose: the walkthroughs he found all ran through Claude Desktop or Cursor, and he wanted the traffic rather than a chat transcript ^[23].

Of the setup step he wrote, "That part is a footnote." ^[3] The schema is where the work is. One property he printed carries a magnifying-glass emoji in its title and `<code>` tags in its description, both written for the Apify Console input form, where they render ^[12]. In a tool definition, he wrote, they are "tokens an agent pays for and markup it has to ignore", and nothing in the path strips them ^[13]. The whole tool definition runs 9,135 characters ^[14]. Divide that by his 23 properties and each one carries about 397 characters into the agent's context ^[1].

One server-side rewrite is good engineering. `prefill` values get appended to the description as `Example values: ["python developer"]`, which turns a Console convenience into an instruction ^[15]. The author calls that "a genuinely good move" ^[16]. It also makes `prefill` documentation. He has about thirty Actors on the Apify Store ^[1], and on a couple of them prefill was a throwaway placeholder; those placeholders are now the example the model imitates ^[17].

The defect he owns is the empty `required` array. "That is my fault, not the server's," he wrote ^[18]. His schema requires nothing, so the definition tells the agent that a call with zero arguments is valid, and the Actor has no useful default search ^[19]. In his words: "An agent that believes the schema will produce an empty run, and the run will succeed while doing so" ^[20].

Treat the 9,135 characters as a measurement of his schema. It is a job-listings Actor that puts Indeed, LinkedIn and company career boards into one schema ^[21], and the figure moves with property count and with how much form copy sits in the descriptions. His threshold for expecting the same result is three: "If your Actor has more than three inputs, the same things will happen to you" ^[22].

The one thing the post withholds is the 24th field, the one the server adds ^[10]. He calls it "the single most important field in the whole exchange" and says he will come back to it ^[11].