when it can access the right external systems. The promise Agents can reason, plan, summarize, write code, and explain ideas in natural language. The wall Without tools, an agent cannot check live data, read private files, update a database, or create a GitHub issue. The integration tax Every new API usually needs custom code, authentication, schemas, testing, errors, and maintenance. Core idea An agent without tools is like a smart person in a locked room: intelligent, but unable to touch the world. MCP changes the integration model: expose capabilities once through a shared protocol, then let compatible AI apps discover and use them safely. Agentcon Mukono at UCU | Agents Without Limits 01
to the outside world. Model Context Protocol is an open standard that lets AI applications connect to external tools, data sources, and workflows through a common interface. AI MCP Weather Files GitHub DB APIs USB-C for AI Different tools can connect through one familiar standard instead of a separate custom integration every time. Agentcon Mukono at UCU | Agents Without Limits 03
architecture. MCP Host The AI application the user interacts with. It manages the experience and coordinates tool use. MCP Client A connector inside the host. Each client maintains a connection to one MCP server. MCP Server A program that exposes useful capabilities such as tools, resources, and prompts. Part Beginner definition Example Host The AI app where the conversation happens. VS Code, Claude Desktop, custom agent app Client The connection manager inside the host. One client for weather, another for GitHub Server The capability provider. Filesystem server, database server, API server Agentcon Mukono at UCU | Agents Without Limits 04
prompts structure repeatable work. T Tools Actions the agent can call. get_weather(location) read_file(path) create_github_issue(title) R Resources Context the agent can read. README files Database schemas Project documents P Prompts Reusable interaction templates. Bug triage Code review Data analysis If the audience remembers only this slide: tools are for actions, resources are for context, prompts are for workflow patterns. Agentcon Mukono at UCU | Agents Without Limits 05
guessing to checking live data. 1 User asks Should I carry an umbrella in Mukono this afternoon? 2 Agent reasons This needs current weather, not only general knowledge. 3 MCP discovers The weather server exposes get_weather(location). 4 Tool runs The server calls a weather API and returns structured results. 5 Agent responds The final answer is grounded in live data. Tool call { "name": "get_weather", "arguments": { "location": "Mukono, Uganda" } } Result The agent can answer with context, confidence, and a practical recommendation. Agentcon Mukono at UCU | Agents Without Limits 06
unlimited access. Without MCP With MCP The model cannot see local files unless the app builds a custom feature. A filesystem MCP server can expose approved folders as resources or tools. Every app implements file access differently. The host discovers the same capability through the protocol. Security is often hidden from the user. Good hosts show the tool call and request approval for sensitive actions. Demo prompt Summarize the README and tell me the main setup steps. Security lesson: restrict scope, validate inputs, show actions, and ask before writes or sensitive reads. Agentcon Mukono at UCU | Agents Without Limits 07
chat assistant into a developer workflow assistant. Understand Read README, issues, branches, and selected files. Triage Search issues, summarize duplicates, and suggest labels. Implement Create a branch, draft a plan, or open a pull request. Stay safe Ask for confirmation before write actions. Demo prompt Find open issues about login errors, summarize top patterns, and draft a new issue with reproduction steps. Agentcon Mukono at UCU | Agents Without Limits 08
through one useful workflow. Level What to build Skill gained Beginner Use an existing MCP server in an AI client. Discovery, permissions, and tool calls. Builder Create a small server with one tool, such as weather or calculator. Tool schemas and clean structured results. Pro Connect business systems, databases, and internal workflows. Secure, observable, production-ready agent capabilities. User request AI host MCP client MCP server API or data Response Start small: one server, one tool, one useful workflow. Agentcon Mukono at UCU | Agents Without Limits 09
access context, and act safely. Tools Let the agent take action. Context Let the agent understand the real situation. Safety Keep users in control. MCP matters because it gives developers one common protocol for connecting AI apps to the outside world. References: modelcontextprotocol.io/introduction, /docs/concepts/architecture, /docs/concepts/tools Agentcon Mukono at UCU | Agents Without Limits 10
into an MCP tool. References: modelcontextprotocol.io/introduction, /docs/concepts/architecture, /docs/concepts/tools Agentcon Mukono at UCU | Agents Without Limits 10
afroboy
smashingmag
codingconduct
aleyda
wayneb77
akosma
irinanazarova
holman
bkeepers
lynnandtonic
jacobtomlinson
keathley
