In this Monkeys Sharing,
Beder Acosta Borges explain to the team how he uses MCP to speed up his development workflow.
- Model Context Protocol (MCP) is a new standard based on JSON RPC that aims to standardize the way AI agents communicate and interact with various data sources, tools, and applications.
- It addresses the “n times n integration problem,” where traditionally, integrating an AI agent with multiple data sources (like GitHub, Gmail, or Slack) required a custom implementation for each unique connection. MCP simplifies this by providing a standardized protocol, meaning you only need to write one server for each data source and one client for the AI agent, allowing for interchangeable use.
- MCP can be conceptualized as a “USB-C port for AI applications,” signifying its role in providing a universal connection. This enables AI assistants not only to fetch content from MCP servers but also to perform actions and modify the server’s status, with a key focus on enriching the context available to the agent.
- A primary goal of the protocol is to enrich the context for AI agents with private, company-specific information such as database schemas, private GitHub repositories, Discord, Sentry, and ClickUp data. This direct access to relevant, internal data allows AI agents to generate significantly better and more accurate code compared to relying on generic prompts or manual copy-pasting.
- MCP facilitates seamless workflow automation by giving AI agents the ability to both retrieve information and execute actions. This capability allows for the automation of repetitive, daily tasks like creating dedicated branches, pushing draft pull requests to GitHub, and updating task statuses in ClickUp.
- The standard promotes a “plug-and-play” integration model, meaning as MCP gains wider adoption, more official or community-maintained servers will become available, simplifying the integration of AI agents with various tools and data sources.
- Security is a core consideration of the protocol. It allows for MCP servers handling sensitive data to run locally, mitigating concerns companies might have about integrating AI agents with private internal information by keeping crucial data within the company’s private network.
- The development process with MCP is iterative: the AI agent can request additional information from MCP servers as it progresses through a task. Users maintain control through manual confirmations before the agent accesses external resources and can provide feedback to correct errors or update coding instructions, for example, by modifying the
copilot_instructions
file. - AI agents can adhere to specific coding standards and architectural conventions by reading them from a
copilot_instructions
markdown file within the repository. This enables the agent to understand project-specific rules, such as how to interact with frameworks like Django, leading to code that aligns with internal development practices. - The possibilities with MCP are considered infinite, as it enables AI agents to make more informed decisions by providing them with rich, relevant context from various internal company tools and data sources, extending beyond code development to areas like billing and invoicing. The speaker mentions they prefer using Cloud 4 for these kinds of tasks.