Tools & Capabilities
Complete reference of all tools, commands, and capabilities available in the Vision system.
Core Tools
File Operations
Read - Read files with line numbers
- Supports images, PDFs, Jupyter notebooks
- Line offset and limit for large files
- Always read before editing
Write - Create or overwrite files
- Must read existing files first
- Supports all text formats
Edit - Exact string replacement
- Preserves indentation
- Must read file first
- Can replace all occurrences
Glob - File pattern matching
- Fast pattern search
- Sorted by modification time
- Works with any codebase size
Grep - Content search (ripgrep)
- Full regex support
- Filter by file type or glob
- Context lines (-A, -B, -C)
- Multiline matching support
Execution Tools
Bash
Terminal command execution with:
- Persistent shell session
- Optional timeout (max 10 minutes)
- Background execution support
- Proper path quoting for spaces
Important: Use for git, ssh, builds, services - NOT for file operations
Task (Agent)
Launch specialized agents for:
- general-purpose: Complex multi-step tasks
- statusline-setup: Configure status line
- output-style-setup: Create output styles
Use when: Multiple rounds of search/analysis needed
Browser Automation
Playwright MCP
Full browser automation:
- Chrome, Firefox, WebKit
- Screenshots and PDFs
- Page interactions
- Form submission
Puppeteer MCP
Chrome automation:
- Headless browsing
- Page navigation
- Element interactions
Protocol: ALWAYS use Chrome (never Safari)
Search & Research
WebSearch
Built-in web search:
- Current events and recent data
- Domain filtering support
- US-only availability
WebFetch
Fetch and analyze web content:
- HTML to markdown conversion
- AI-powered analysis
- 15-minute cache
DuckDuckGo MCP
Privacy-focused web search
Wikipedia MCP
Article search and retrieval
YouTube Transcript MCP
Video transcript extraction
Development Tools
SQLite MCP
Database operations:
- Schema inspection
- Query execution
- Data manipulation
AST-Grep MCP
Code structure search:
- AST-based pattern matching
- Language-aware search
Fetch MCP
HTTP/HTTPS requests:
- API testing
- Web scraping
GitHub (via docker-mcp-gateway)
Repository operations:
- Clone, pull, push
- Issue/PR management
- Code search
Documentation
Cloudflare Docs MCP
Complete Cloudflare reference:
- Workers, Pages, R2, D1, KV
- AI, Zero Trust, CDN, DNS
JetBrains MCP
IDE documentation:
- IntelliJ, PyCharm, WebStorm
Atlas Docs MCP
MongoDB Atlas guides
DockerHub MCP
Container documentation
Communication
Slack MCP
Direct #qc channel access:
- Post messages
- Read conversations
- Team: T09L15F1KDG
Error Tracking
Flare MCP (HTTP)
Production error monitoring:
- Stack traces
- User context
- Environment data
- Error frequency
Protocol: Check Flare FIRST for production errors
System Access
Desktop Commander MCP
macOS filesystem control:
- Full root access (via Docker volume mount)
- Directory operations
- File management
AppleScript
Direct app control for 40+ apps:
- Chrome, Mail, Reminders
- Spotify, iMessage
- Terminal operations
AI & Media
HuggingFace Spaces MCP
- shuttle-3.1-aesthetic: Image generation
- whisper-large-v3-turbo: Audio transcription
- QVQ-72B-preview: Vision model
- OmniParser: Document parsing
Infrastructure Management
DigitalOcean MCP
Complete DO API:
- Droplet management
- DNS and domains
- Firewalls, load balancers
- Monitoring, alerts
- Billing, SSH keys
Cloudflare Workers MCP
Workers development:
- Deployment
- Debugging
- Analytics
Tool Decision Matrix
| Task | Use | NOT |
|---|---|---|
| Read files | Read tool | cat, head, tail |
| Edit files | Edit tool | sed, awk, vim |
| Write files | Write tool | echo, cat |
| Search content | Grep tool | grep command |
| Find files | Glob tool | find, ls |
| Execute commands | Bash | - |
| Browser testing | Playwright | curl |
| Web content | WebFetch | curl |
| Communication | Direct text | echo |
Wrong tool = waste money
Capability Limits
What Vision CAN Do
- Read/write any file on system
- Execute any bash command
- Full browser automation
- API calls and web scraping
- Database operations
- Server management (SSH)
- Git operations
- Docker management
- AppleScript automation
What Vision CANNOT Do
- Interactive terminal (no -i flags)
- Real-time GUI interaction
- Direct hardware access
- Kernel-level operations
- Break out of security boundaries
Performance Considerations
Parallel Tool Calls:
- Use multiple tools in one message when independent
- Reduces round-trip time
- Example: Read multiple files simultaneously
Background Execution:
- Long-running bash commands
- Use BashOutput to check progress
- Don't block on slow operations
Cache Utilization:
- WebFetch has 15-minute cache
- Read files once, reference multiple times
Total Capabilities: 21 MCP servers + native Claude Code tools Status: Fully operational (5/21 MCP active, Docker offline)