From atelier
Add a new MCP server to the project mid-project. Wraps the capability-management procedure: external MCP search, ADR draft, .mcp.json update, user install instructions, and capability-log entry. User approval is mandatory regardless of involvement level. Use when an authorized agent identifies a missing external integration.
How this skill is triggered — by the user, by Claude, or both
Slash command
/atelier:add-mcpThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are the Conductor (or a delegated authorized agent) orchestrating MCP server addition. MCP servers add privileged surface area (network access, secrets, tool execution), so this flow has stricter gates than skill creation.
You are the Conductor (or a delegated authorized agent) orchestrating MCP server addition. MCP servers add privileged surface area (network access, secrets, tool execution), so this flow has stricter gates than skill creation.
Only these agents may invoke this skill:
chief-ai-officersoftware-architecttech-lead (infrastructure additions only)All other agents must escalate.
/atelier:add-mcp <mcp-name> "<reason>"
<mcp-name> — the MCP server identifier (e.g., @modelcontextprotocol/slack).<reason> — one sentence explaining the capability gap.Per docs/process/capability-management.md. Especially check:
.mcp.json entries — is something already covering this need?Record the audit evidence.
Use WebFetch / WebSearch to find candidates:
mcp-server.Evaluate each: maintenance status, license, security review, auth model.
Create docs/ssot/decisions/adr-NNN-mcp-<name>.md from the template in docs/README.md#ssot-schema. Required content:
Show the user:
Ask:
"Approve adding
<mcp-name>to this project's MCP servers?"
Do NOT proceed without explicit approval, regardless of involvement level (Fully Autonomous still requires approval here).
.mcp.jsonOnce approved, add the MCP entry to the project's .mcp.json. Reference secrets by environment variable name only — the user installs and configures the actual secrets in their MCP configuration, not in repo files.
Print the exact commands the user must run, e.g.:
# Install the MCP server
npm install -g @modelcontextprotocol/slack
# Configure auth (user fills with real token)
export SLACK_BOT_TOKEN="<your-token>"
# Reload Claude Code plugins
/reload-plugins
The agent does NOT run install commands itself — MCP installation is a user-privileged action.
Append to docs/agents/capability-log.md:
| <YYYY-MM-DD> | mcp | <mcp-name> | <invoking-agent> | <reuse audit summary> | <one-sentence justification> | adr-NNN |
Once user has installed and reloaded, update the ADR status: Proposed → Accepted.
npm install / equivalent on the user's behalf for MCP servers.If more than 2 MCP additions occur within a single milestone, Chief AI Officer is automatically re-invoked at the next milestone-checkpoint to review whether the project's external dependency footprint has grown unhealthily.
docs/process/capability-management.mddocs/README.md#ssot-schemaProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Searches MemPalace before answering questions about past work, people, projects, or prior decisions. Returns verbatim stored content instead of guessing from model memory.
npx claudepluginhub dudgns0908/atelier --plugin atelier