mcp-management_skill

This skill enables managing MCP servers, discovering capabilities, and executing tools to keep context clean and improve task-driven automation.

Python

2

GitHub Stars

2

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill mamba-mental/agent-skill-manager --skill mcp-management

README.md5.6 KB
SKILL.md6.3 KB

Overview

This skill manages Model Context Protocol (MCP) servers to discover, analyze, and invoke tools, prompts, and resources across multiple configured MCP endpoints. It provides configuration handling, capability discovery, intelligent tool selection, and several execution patterns to keep the main agent context clean. Persistent catalogs accelerate repeated use and support offline analysis.

How this skill works

The skill reads MCP server configuration, queries each server for tools/prompts/resources, and aggregates results into a persistent JSON catalog. An LLM-driven analyzer inspects the catalog to match tools to task intents, while execution can run via Gemini CLI, direct CLI scripts, or a subagent fallback that isolates context. Results and error handling are returned with server provenance for routing and audit.

When to use it

Discover available tools, prompts, or resources exposed by MCP servers
Analyze which MCP tools best fit a specific task or intent
Execute MCP tools programmatically while preserving the main agent context
Integrate or debug an MCP client implementation across multiple servers
Coordinate tool orchestration across different MCP servers with clear provenance

Best practices

Prefer Gemini CLI integration as the primary execution path for natural-language invocation and automatic discovery
Keep .claude/.mcp.json synchronized with any orchestration settings and optionally symlink to .gemini/settings.json for CLI tools
Use the persistent assets/tools.json catalog for offline analysis and to let LLMs select tools without live server calls
Fallback to direct scripts or a subagent when Gemini CLI is unavailable to avoid polluting the main context
Include server identifiers with every tool result to ensure correct routing and error tracing

Example use cases

List and persist all MCP tools from multiple servers to assets/tools.json for audit or review
Ask an LLM to pick the best tool(s) for data extraction, then execute the chosen tool via Gemini CLI
Invoke a specific tool with precise arguments using the CLI script call-tool <server> <tool> <json> for programmatic workflows
Run an mcp-manager subagent to discover and execute tools while keeping the primary agent conversation minimal
Orchestrate a multi-server workflow where different tools on separate servers handle discrete steps of a pipeline

FAQ

Three patterns: Gemini CLI (primary), direct CLI scripts (secondary), and an mcp-manager subagent (fallback) to preserve main context.

Where are discovered tools stored?

Discovered capabilities are saved to assets/tools.json (complete schemas and server provenance) for fast reference and offline use.