Understanding the
AI Tooling Ecosystem
Learn about MCPs and Skills - the building blocks of our internal AI tooling platform. Understand why running locally matters and how to leverage these tools effectively.
Why We Run AI Tools Locally
Our AI tooling platform is built on a local-first philosophy. Here's why that matters for enterprise teams at Deloitte.
Data Never Leaves Your Machine
Sensitive client data, credentials, and proprietary information stay on your local environment. Nothing is transmitted to external servers.
Enterprise Security Compliance
Meet SOC 2, HIPAA, and client-specific security requirements by keeping all processing within your controlled environment.
Lightning Fast Execution
No network latency for tool calls. Local MCP servers respond in milliseconds, making AI interactions feel instant.
Full Transparency & Auditability
See exactly what tools are being called, what data they access, and what they return. Complete audit trail on your machine.
Offline Capable
Many tools work without internet. Query local databases, analyze local files, and generate content even without connectivity.
No Vendor Lock-in
MCP is an open protocol. Your tools work with Claude, GPT, Gemini, and any future AI model that supports the standard.
How It All Fits Together
AI Model
Claude, GPT-4, Gemini
Processes requests and uses tools
Local Runtime
Your Machine
MCPs and Skills run here
Enterprise Services
APIs, Databases, SaaS
MCPs bridge to external systems
The key insight:Your data flows between the AI model and your local machine. Sensitive information never touches third-party servers you don't control.
MCP Servers
Model Context Protocol
MCP Servers are lightweight programs that run on your local machine and extend your AI assistant with new capabilities. They follow the open Model Context Protocol standard to provide tools, resources, and prompts to AI models.
How It Works
Your AI client (like Claude Desktop) connects to MCP servers via stdio or HTTP
The server advertises available tools, resources, and prompt templates
When you ask your AI something that needs a tool, it calls the MCP server
The server executes locally and returns results to the AI
Use Cases at Deloitte
Query databases without exposing credentials to the cloud
Access local file systems and project structures
Interact with internal APIs and services securely
Provide real-time data from enterprise systems
Example: SQL Query MCP
// claude_desktop_config.json
{
"mcpServers": {
"deloitte-sql": {
"command": "npx",
"args": ["-y", "@deloitte-mcp/sql-analyst"],
"env": {
"DB_CONNECTION": "your-connection-string"
}
}
}
}Skills
Domain Expertise Modules
Skills are curated packages of domain expertise that enhance AI performance in specific areas. They combine prompt engineering, few-shot examples, and specialized knowledge to make AI consistently excellent at particular tasks.
How It Works
Skills package domain knowledge into reusable prompt templates
They include few-shot examples that guide AI behavior
Skills can be composed together for complex workflows
They run entirely within your AI context - no external calls needed
Use Cases at Deloitte
Financial modeling and analysis with Deloitte methodologies
Legal document review following compliance frameworks
Code review aligned with enterprise quality standards
Risk assessment using proven analytical approaches
Example: Financial Analysis Skill
// Skill definition
{
"name": "financial-modeling-pro",
"domain": "Financial Analysis",
"capabilities": [
"DCF modeling",
"Sensitivity analysis",
"Scenario planning"
],
"promptTemplate": "You are a senior financial analyst..."
}Getting Started
Three steps to supercharge your AI workflows
Browse the Marketplace
Explore our curated collection of MCPs and Skills built by Deloitte teams.
Install Locally
Each tool comes with a one-line install command. Run it on your machine and configure your AI client.
Contribute Back
Built something useful? Submit it for review and earn points on the leaderboard.
Frequently Asked Questions
Do I need to be a developer to use these tools?
Not at all. Most tools install with a single command and configure through simple JSON files. Skills require zero technical setup.
Is my data safe when using MCPs locally?
Yes. MCPs run entirely on your machine. Your data never leaves your local environment unless you explicitly configure an MCP to connect to an external service.
Can I use multiple tools together?
Absolutely. That is the power of the ecosystem. You can install multiple MCPs alongside Skills, and Skills can be composed to handle complex domain tasks.
What AI models are supported?
MCP is an open protocol supported by Claude, and increasingly by other AI models. Skills work with any AI model.
How do I get my tool approved for the marketplace?
Submit your tool through the publish wizard. Our team reviews for security, quality, and documentation completeness. Most tools are reviewed within 48 hours.
What happens if a tool stops working after an update?
All tools are versioned. You can pin to a specific version and update on your own schedule. The marketplace shows compatibility information and changelogs.