Making MCP cheaper via CLI
kanyilmaz.me
February 25, 2026
3 min read
Summary
MCP agents incur high costs due to the inclusion of an extensive tool catalog in JSON Schema format. A CLI generated from an MCP server using CLIHub performs the same functions at a significantly reduced cost.
Key Takeaways
- CLI reduces token usage by approximately 94% compared to MCP by using a lightweight skill listing instead of pre-loading full tool schemas.
- The session start cost for MCP is around 15,540 tokens, while CLI only requires about 300 tokens.
- Anthropic's Tool Search, while more efficient than MCP, still incurs higher costs than CLI, making CLI the cheaper option for tool discovery and execution.
- The author created CLIHub, a directory of CLIs for agent use, and open-sourced a converter to generate CLIs from MCPs with a single command.
Community Sentiment
MixedPositives
- The CLI approach offers practical benefits for token reduction, which can enhance efficiency in AI interactions by minimizing unnecessary context bloat.
- Existing paradigms like Skills effectively mitigate context bloat and selection overhead, showcasing a more structured approach to tool discovery in AI applications.
Concerns
- MCP tools contribute significantly to context bloat, as they require sending extensive tool catalogs with every request, which can hinder performance.
- The claim that MCP tools need to dump the entire tool catalog into the conversation is outdated, as more advanced clients like Claude Code have improved this process.
Related Articles
Apideck CLI – An AI-agent interface with much lower context consumption than MCP
Mar 16, 2026
MCP is dead; long live MCP
Mar 14, 2026
When does MCP make sense vs CLI?
Mar 1, 2026
Stop Burning Your Context Window – How We Cut MCP Output by 98% in Claude Code
Feb 28, 2026
You need to rewrite your CLI for AI agents
Mar 4, 2026
Source
kanyilmaz.me
Published
February 25, 2026
Reading Time
3 minutes
Relevance Score
63/100
Why It Matters
This page is optimized for focused reading: quick context up top, a clean summary block, and a direct path to the original source when you want the full story.