I launched an MCP server on Product Hunt today. Its real user is the agent.

Description and use

Agent scripts became more efficient thanks to updated tool descriptions that include purpose, cost, and alternatives, reducing duplicate requests and costs.

Problem

Agents repeatedly called sub‑optimal web tools, wasting credits and time.

Outcome

Agents now use fewer calls, pay less per token, and handle bot walls gracefully, improving efficiency and cost control.

Workflow

• Parse server call logs to identify redundant chains
• Rewrite tool descriptions to include cost and alternative usage
• Return markdown output to reduce token usage
• Report bot walls as blocks with vendor info
• Use local Ollama model by default for LLM tasks
• Add CrawlForge MCP to Claude via CLI command

Evidence

21% of calls repeated an identical call within 30 minutes.

Source

Use this to read one page - markdown by default, plus any of "html", "rawHtml", "text", "links", "metadata", ... all from one fetch.

Source

Raw HTML is mostly navigation, scripts and attributes the model has to read and then ignore.

Source