Claude Code CLI

CLI Configuration

Settings, permissions, MCP servers, agents, and skills — how Claude Code is wired for this infrastructure

2
Config Levels
15
Skills
5
Agents
71
MCP Tools
174
Pre-approved Permissions
Structure

Two Configuration Levels

User-level is personal. Project-level is infrastructure-wide.
User Level
~/.claude/ ├─ settings.json # global config ├─ CLAUDE.md # user identity ├─ agents/ # pm agent ├─ skills/ # 7 user skills └─ commands/ # 6 slash cmds

Personal preferences, general-purpose skills, and the PM agent. Applies across all projects on this machine.

Project Level
~/projects/.claude/ ├─ settings.local.json # 174 perms ├─ agents/ # 4 specialists ├─ skills/ # 8 infra skills └─ commands/ # 4 project cmds

Infrastructure-wide skills, specialist agents, and pre-approved permissions for autonomous operation across all services.

Settings

Key Configuration Settings

The most important knobs that shape how Claude Code behaves

~/.claude/settings.json — Global

alwaysThinkingEnabled
true — Extended reasoning is on for every request. Claude thinks through problems before responding.
enableAllProjectMcpServers
true — All MCP servers defined in project config are automatically available without extra prompting.
permissions.deny
deny Read(/projects/administrator/projects/claude) — Prevents reading the raw Claude binary path.

~/projects/.claude/settings.local.json — Pre-approved Permissions

85+
Bash commands
docker, git, psql...
7
Read paths
secrets, projects...
8
WebFetch domains
github, docs...

Pre-approved permissions let Claude work autonomously without interrupting you for confirmation on routine infrastructure tasks.

MCP

MCP Integration

Model Context Protocol — Claude's bridge to external tools and databases
A single MCP server (code-executor) runs inside Docker. It proxies requests to 9 backend servers, giving Claude access to 71 tools — databases, object storage, browser automation, workflows, and market data.
DB
postgres
Query databases, inspect schemas, run migrations
9 tools
FS
filesystem
File operations within /workspace
9 tools
S3
minio
S3-compatible object storage, file sharing
9 tools
BR
playwright
Browser automation, screenshots, web scraping
6 tools
WF
n8n
Trigger and manage automation workflows
6 tools
TS
timescaledb
Time-series database for metrics and trading data
6 tools
MEM
memory + arangodb
Knowledge graph and persistent memory storage
16 tools
IB
ib (Interactive Brokers)
Live and paper trading market data
10 tools
Agents

Specialized Agents

Sub-agents invoked via the Task tool — each with focused expertise and its own context
PM — Project Manager (user-level)
Orchestrates multi-step projects. Breaks work into phases, assigns tasks to specialist agents, tracks progress across sessions. The starting point for any complex request.
Architect
System design decisions: technology stack selection, service integration patterns, performance trade-offs, scalability planning. Produces architecture specifications for the Developer to implement.
Developer
Hands-on implementation: writes code, creates Dockerfiles, deploys services, debugs containers, manages databases. Follows the Architect's design and Security's configuration exactly.
Security
Network topology, Traefik routing, Keycloak OAuth2 clients, firewall rules, SSL/TLS certificates. Owns all network-layer decisions that the Developer deploys but does not design.
QA
Quality assurance and testing. Runs tests via GitLab CI pipelines, validates deployments, checks service health, and verifies that implementations meet specifications.

Each agent reads the shared CLAUDE.md agent-docs, plus its own specialized context. Agents communicate via escalation files and the Matrix chat network.

Summary

Configuration at a Glance

15
Skills
(7 user + 8 infra)
5
Agents
(PM, Arch, Dev, Sec, QA)
71
MCP Tools
across 9 servers
174
Pre-approved
permissions
10
Slash commands
(6 user + 4 project)
2
Config levels
user + project

Design intent: Two config levels keep personal preferences separate from infrastructure standards. Pre-approved permissions enable autonomous operation without constant prompts. MCP tools extend Claude's reach into every layer of the stack.

Claude Code CLI Configuration • ai-servicers.com