The Model Context Protocol (MCP) is an open protocol for connecting AI models to external data sources and tools. It provides a standardized way for coding agents — such as GitHub Copilot, Claude Code, and other AI assistants — to discover and interact with structured context beyond what is in their training data.
MCP defines three core primitives:
- Prompts — Reusable prompt templates that a server can serve to an agent on demand.
- Resources — Structured data (documents, schemas, data dictionaries) that an agent can query.
- Tools — Executable capabilities (validation, computation, generation) that an agent can invoke.
For more details, see the MCP specification and the MCP GitHub repository.
Frontier Analytics is built around prompt cards, data schemas, and structured workflows that analysts paste into coding agents. Today, this is a manual process: you open a prompt card, copy the text, and paste it into your agent along with context about your data.
MCP could change this by enabling coding agents to directly access the Frontier Analytics library:
- Prompt cards served on demand. Instead of copying a prompt card from a Markdown file, a coding agent could request the appropriate prompt from an MCP server and receive it pre-formatted and ready to execute.
- Schema documentation as queryable resources. A coding agent could ask "What columns are in the person query export?" and receive a structured answer from the schema docs — no manual lookup needed.
- Validation and computation as tools. A coding agent could validate a CSV against the expected schema, compute standard metrics, or generate a dashboard template by calling tools exposed through MCP.
This would reduce friction, improve consistency, and allow the prompt library to scale without requiring analysts to manually navigate the repository.
This is forward-looking, not current functionality.
MCP integration for Frontier Analytics is conceptual. No MCP server has been implemented. The contents of this folder are documentation and sample configurations that explore how MCP could enhance the Frontier Analytics workflow in the future.
The current file-based structure of prompts, schemas, and starter kits is designed to be MCP-compatible — the structured, consistent format of prompt cards and schema docs means they could be served through an MCP server with minimal transformation when the time comes.
| File | Description |
|---|---|
| frontier-mcp-concepts.md | Conceptual document explaining how MCP primitives map to Frontier Analytics components, potential use cases, and implementation considerations |
| sample-config/example.mcp.json | A sample MCP client configuration file showing what a Frontier Analytics MCP server integration might look like |
The example.mcp.json file in sample-config/ is a conceptual example of an MCP client configuration. Since JSON does not support comments, the field names and values are written to be self-documenting. It shows how a coding agent's MCP client might be configured to connect to a hypothetical Frontier Analytics MCP server.
This file is not functional — there is no server to connect to. It is provided as a reference for what an MCP integration could look like.
| Today (manual) | Future with MCP |
|---|---|
| Analyst browses the prompts directory to find the right card | Agent queries the MCP server for prompts matching the analyst's goal |
| Analyst copies prompt text and pastes it into the agent | Agent receives the prompt directly from the server, pre-formatted |
| Analyst manually checks column names against schema docs | Agent queries the schema resource to validate columns automatically |
| Analyst reviews common failure modes after the fact | Agent checks failure modes proactively before executing |
| Analyst manually combines outputs from multiple prompt cards | Agent orchestrates a multi-step workflow using the starter kit as a plan |