luum-home/luum-cognitive-os/add-mcp

Trust tier 3
  • community
  • ext:[{"type":"config","description":"Configuration overrides"}]
Source check pending

Checking that the SKILL.md file is still in the repository.

'Use when you need this Cognitive OS skill: Step-by-step guide for integrating

Install

npx vskill@latest install luum-home/luum-cognitive-os/add-mcp

Semi-extensibleMentions customization, but needs skill-specific knowledge. Read the extensible skills standard

SKILL.md

Add MCP Server

Procedure for integrating a new MCP server so its tools are available to agents.

Trigger

When you need to give Claude access to a new external capability via the Model Context Protocol.

Inputs

  • MCP server name: identifier for the server (e.g., my-service)
  • Tools provided: list of tool names the server exposes
  • Install command: how to install the server (npm package, binary, npx, etc.)
  • Activation condition: always-on, or opt-in via env var

Steps

1. Install the MCP server

Install the server binary or package:

bash
# npm package (global install)
npm install -g @scope/mcp-server-name
# npx (zero-install, runs on demand)
npx -y @scope/mcp-server-name
# Go binary
go install github.com/org/mcp-server@latest
# Homebrew
brew install org/tap/mcp-server

If no API key or credentials are needed, skip to step 2. If credentials are required:

  • Store in environment variable (never hardcoded)
  • Reference as ${ENV_VAR_NAME} in the config (see step 2)
  • Document the env var name in step 4
2. Register in .claude/settings.json under mcpServers

Add the server configuration:

json
{
"mcpServers": {
"{mcp-server-name}": {
"command": "npx",
"args": ["-y", "@scope/mcp-server-name"],
"env": {
"API_KEY": "${MY_SERVICE_API_KEY}"
}
}
}
}

For a binary (not npx):

json
{
"mcpServers": {
"{mcp-server-name}": {
"command": "mcp-server-binary",
"args": ["--flag", "value"]
}
}
}

Note: Register in .claude/settings.json (project-level) for project-scoped MCP servers, or in ~/.claude/settings.json (user-level) for servers available in all projects.

3. Document in packages/ecosystem-tools/rules/ecosystem-tools.md

Add a new entry in the "Integrated Tools" section:

markdown
### {Service Name} — {Brief Purpose}
| Property | Value |
|----------|-------|
| Purpose | {What the MCP server does} |
| Install | `{install command}` |
| Required | No (optional, graceful skip if missing) |
| Scope | {What it operates on} |
| License | {License type} |
| Status | **ADOPT** / **EVALUATE** / **WATCH** |
**Usage examples**:

Example invocation

**Tools provided**:
| Tool | Purpose |
|------|---------|
| `tool_name_1` | Description |
| `tool_name_2` | Description |
4. Add graceful degradation (if the server is optional)

If the MCP server is optional (not required for core OS function), document the fallback behavior. For hooks that use the server's tools, add a silent skip guard:

bash
# In any hook or script that calls the MCP server
if ! command -v mcp-server-binary &>/dev/null; then
# Server not installed — skip silently
exit 0
fi

For optional servers, set Required: No in the ecosystem-tools.md entry.

5. Document required environment variables

If the server needs API keys or credentials, document them:

  • Add to docs/environment-variables.md (if the file exists) or note in the ecosystem-tools.md entry
  • Add example to .env.example if the project has one:

``bash # Required for {MCP server name} integration MY_SERVICE_API_KEY=your_key_here ``

  • Follow rules/credential-management.md: never hardcode values, always use ${ENV_VAR} references
6. Verify the integration

Restart Claude Code and verify the server loads:

bash
# Check Claude Code loads the MCP server without errors
# (Restart is required for settings.json changes to take effect)

After restart, the tools from the MCP server should be available. Test with a simple invocation:

# In Claude Code: try calling one of the documented tools

MCP Server Types Reference

TypeExamplesConfig notes
Memory/knowledgeEngram, Context7Usually plugin-based
DocumentationContext7Per-library doc lookup
API connectorsGitHub, Jira, SlackRequire API credentials
Development toolsE2B sandbox, mcp-aguaraRequire install
File managementGoogle Drive, NotionRequire OAuth setup

Output: Configured MCP Server

  • .claude/settings.json — updated mcpServers section
  • packages/ecosystem-tools/rules/ecosystem-tools.md — documented entry
  • Graceful degradation added (if optional server)
  • Environment variables documented

Success Criteria

  • [ ] Server listed in .claude/settings.json under mcpServers
  • [ ] grep "{mcp-server-name}" packages/ecosystem-tools/rules/ecosystem-tools.md returns a match
  • [ ] If credentials needed: ${ENV_VAR} syntax used (not hardcoded values)
  • [ ] After Claude Code restart: no errors about the MCP server in session start output
  • [ ] If optional: graceful skip guard present in any hook that uses the server

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Extension points

  • configConfiguration overrides

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/luum-home/luum-cognitive-os/add-mcp/badge)](https://verified-skill.com/skills/luum-home/luum-cognitive-os/add-mcp)

Something wrong with this skill or its page? Tell us.