MCP Optimizer MCP 伺服器 | PlugMCP跳到正文mcp-optimizer
Marketplace-ready Claude Code plugin for MCP diagnostics, usage analysis, project scoping, and on-demand skill conversion.
mcp-optimizer helps you answer four separate questions:
- Are my MCP servers healthy?
- Which servers are actually wasting tokens?
- Should I keep MCP but scope it to this project?
Comprehensive MCP optimization toolkit for Claude Code — diagnose, health-check, optimize, and convert MCP servers.
Should I convert a heavy MCP server into local on-demand skills?
- explicit plugin commands in
commands/
- matching bundled Skills in
skills/
Each /mcp-optimizer:* command is a thin entrypoint that delegates to the corresponding bundled Skill, so the workflow logic stays in one place.
Why This Exists
When an MCP server is connected to Claude Code, its tool schemas are loaded into every conversation whether you use them or not.
That creates three common problems:
- Large MCP schemas waste tokens even when idle.
- Broken or duplicate servers pollute context and slow down sessions.
- Global MCP configs are often broader than a single project needs.
Installation
From GitHub (Recommended)
-
Add this repository as a marketplace.
/plugin marketplace add choam2426/mcp-optimizer
-
Install the plugin.
/plugin install mcp-optimizer
-
Restart Claude Code if prompted, then verify installation.
/help
From Another Marketplace
If mcp-optimizer is included in an existing marketplace:
/plugin marketplace add your-org/claude-plugins
/plugin install mcp-optimizer@your-org
Components At a Glance
Recommended Workflow
-
Start with diagnostics:
/mcp-optimizer:mcp-doctor
-
Once servers are healthy, measure actual waste:
/mcp-optimizer:mcp-audit
-
Choose the execution path that fits your goal:
-
Keep MCP, but only for this project:
/mcp-optimizer:mcp-optimize
-
Convert a heavy server into on-demand local skills:
/mcp-optimizer:mcp-to-skills npx @linear/mcp-server
The important distinction is:
mcp-doctor is a diagnostic command, not an optimization command.
mcp-audit is an analysis command that tells you where waste comes from.
mcp-optimize and mcp-to-skills are the actual optimization paths.
Command Reference
Diagnostics: /mcp-optimizer:mcp-doctor
Use mcp-doctor when you want to inspect MCP server health before changing anything.
- tests configured MCP servers
- measures response time
- finds duplicate tool names across servers
- flags missing or suspicious credentials
- suggests concrete remediation steps
/mcp-optimizer:mcp-doctor [--server <name>] [--fix]
- Runs as a read-only health check.
- Does not modify your config.
- If
--fix is present, it suggests commands or follow-up actions, but still does not edit anything automatically.
- health report by server
- response times
- duplicate tool findings
- missing credential warnings
- recommended next step, usually
/mcp-optimizer:mcp-audit after issues are resolved
/mcp-optimizer:mcp-doctor
/mcp-optimizer:mcp-doctor --server github
/mcp-optimizer:mcp-doctor --fix
Analysis: /mcp-optimizer:mcp-audit
Use mcp-audit when you want evidence about token waste before deciding what to keep, remove, scope, or convert.
- reads Claude Code session history
- estimates per-server schema overhead
- shows which tools were used vs never used
- ranks optimization opportunities
- suggests conversion commands for high-waste servers
/mcp-optimizer:mcp-audit [--scope project|all]
- If no scope is provided,
project scope is used.
- The command only analyzes available session history. It does not modify config or generated skills.
- ranked token waste by MCP server
- total estimated waste
- unused tool ratios
- recommendations for
/mcp-optimizer:mcp-optimize and /mcp-optimizer:mcp-to-skills
/mcp-optimizer:mcp-audit
/mcp-optimizer:mcp-audit --scope project
/mcp-optimizer:mcp-audit --scope all
Optimization: /mcp-optimizer:mcp-optimize
Use mcp-optimize when you still want MCP, but not every global server should load in this project.
- analyzes which servers look relevant to the current project
- proposes a reduced project-local
.mcp.json
- estimates token savings
- writes the file only after confirmation
/mcp-optimizer:mcp-optimize [--dry-run] [--min-sessions <N>]
- Treats optimization as a plan-first workflow.
- Shows the proposed
.mcp.json before writing.
- Writes only the project-local
.mcp.json.
- Never modifies global Claude config such as
~/.claude.json.
- keep/remove recommendation per MCP server
- estimated savings per session
- full proposed
.mcp.json
- confirmation step before writing
/mcp-optimizer:mcp-optimize
/mcp-optimizer:mcp-optimize --dry-run
/mcp-optimizer:mcp-optimize --min-sessions 10
Conversion: /mcp-optimizer:mcp-to-skills
Use mcp-to-skills when a server has many tools but you only need a few of them on demand.
- inspects the target MCP server
- reads tool schemas
- generates one local Claude Code skill per tool
- defaults to proxy mode unless native mode is clearly better
/mcp-optimizer:mcp-to-skills <server-command>
- Requires a concrete MCP server command.
- Writes generated local skills under
.claude/skills/.
- Prefers proxy mode by default.
- detected service name
- generated skill directories
- per-tool descriptions
- example natural-language requests for the generated skills
/mcp-optimizer:mcp-to-skills npx @linear/mcp-server
/mcp-optimizer:mcp-to-skills uvx mcp-server-gitlab
Command Vs Skill
This plugin exposes the same workflows in two forms:
commands/: explicit user-invoked commands such as /mcp-optimizer:mcp-audit
skills/: bundled Skills Claude can discover automatically from natural-language requests
- the command is the public explicit entrypoint
- the matching Skill is the canonical workflow implementation
- the command delegates to the matching Skill instead of duplicating instructions
That means you can either:
- invoke the command directly, or
- ask in natural language and let Claude discover the bundled Skill
Generated Local Skills
/mcp-optimizer:mcp-to-skills creates local skills under .claude/skills/.
- Generated local skills are not namespaced as
/mcp-optimizer:*.
- They are separate local skills for your project or working tree.
- Claude should discover them when the user asks for that workflow in natural language.
Example request after conversion:
List open Linear issues for the Engineering team
Example Scenarios
Requirements And Notes
- Python 3.10+
- Claude Code installed and running
- MCP server environment variables must already be available when using
mcp-doctor or mcp-to-skills
mcp-audit depends on available Claude Code session history
mcp-optimize only writes a project-local .mcp.json
License
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
如何安裝 MCP Optimizer?
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
MCP Optimizer 需要 API 金鑰嗎?
這筆記錄未列出必要的環境變數。
MCP Optimizer 使用哪種傳輸方式?
MCP Optimizer 透過 stdio 在本機執行。
MCP Optimizer 上次檢查是什麼時候?
上次檢查時間:2026-10-28。 Comprehensive MCP optimization toolkit for Claude Code — diagnose, health-check, optimize, and convert MCP servers.
azure-computeAzure VM/VMSS router. WHEN: create / provision / deploy / spin-up VM, recommend VM size, compare VM pricing, VMSS, scale set, autoscale, burstable, lightweight server, website, backend, GPU, machine l 工具
0 / 10