Skip to content
Back to skills

562 Mcp Management F64a973a

BSecurity

When I finished reading [Anthropic’s “Code execution with MCP” article](https://www.anthropic.com/engineering/code-execution-with-mcp), a sudden idea flashed in my mind As many people may already know, subagents have their own context windows, while using MCP as it currently does will bloat the main context (anyone who has used Chrome Devtools MCP or Playwright MCP knows how much their tools consume context from the start) So then: **why don’t we load all MCP into the subagent’s context?** *I...

  • 9 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 11, 2026
ai-agentsgogit

Works with

  • cli
  • mcp

Security analysis

B75/100
  • criticalUses Unicode tricks to hide instructions — documented evasion technique in ToxicSkills

Pro shows the line behind each finding and how to fix it

Scanned October 11, 2026

npx -y skills add tools-only/X-Skills --skill 562-mcp_management_f64a973a --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of 562 Mcp Management F64a973a?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for 562 Mcp Management F64a973a
[![Security: B — Skills Directory](https://www.skillsdirectory.com/api/skills/tools-only-562-mcp-management-f64a973a/badge)](https://www.skillsdirectory.com/skills/tools-only-562-mcp-management-f64a973a)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

SKILL.md
# No more context bloat with "mcp-manager" subagent + "mcp-management" skills!

When I finished reading [Anthropic’s “Code execution with MCP” article](https://www.anthropic.com/engineering/code-execution-with-mcp), a sudden idea flashed in my mind

As many people may already know, subagents have their own context windows, while using MCP as it currently does will bloat the main context (anyone who has used Chrome Devtools MCP or Playwright MCP knows how much their tools consume context from the start)

So then: **why don’t we load all MCP into the subagent’s context?**

*I tested it immediately...*

The idea is very simple: **“mcp-manager” subagent + “mcp-management” skills**

1/ **“mcp-management” skills** will have script snippets to initialize MCP Client from “.claude/.mcp.json” (I move the “.mcp.json” file here so the main agent doesn’t load them into context from the start)

2/ **“mcp-manager” subagent** is equipped with **“mcp-management” skills**

Whenever needing to call a tool -> summon “mcp-manager” subagent -> activate “mcp-management” skills -> load MCP servers -> subagent receives list of tools & analyzes to select the tool to use -> call tool & receive result -> return it back to main agent

***Voilà!***

Main context stays pristine and clean even if you use 80 MCP servers 👌

Look at this image and you’ll understand better:

![No more context bloat with "mcp-manager" subagent + "mcp-management" skills](https://github.com/mrgoonie/claudekit-skills/blob/main/assets/mcp-management.jpeg?raw=true)

Actually, after that I upgraded it a bit, because processing such a large number of MCP servers tools, while not polluting the main context, still… consumes tokens, leading to quickly hitting the limit.

So I transferred that MCP processing part to… `gemini-cli` 😂​​​​​​​​​​​​​​​​

I think Anthropic should adopt this approach as default, oc without the "gemini" part.

Apparently it's already available in [ClaudeKit.cc](https://claudekit.cc) 😁

If you find this useful, please support my project:
[![ClaudeKit Agent Skills](https://github.com/mrgoonie/claudekit-skills/blob/main/claudekit.png?raw=true)](https://claudekit.cc)

Cheers,
@mrgoonie

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.