
MCPA (Model Context Protocol Associate) is the first certification for the Model Context Protocol (MCP), the open standard that connects AI applications and agents to tools, data and other systems. The Agentic AI Foundation launched it with Linux Foundation Training and Certification. It is a beginner-level, multiple-choice exam on how MCP works, how hosts, clients and servers interact, how tool calls run and fail, and how to keep MCP deployments secure. For freshers and developers in Hyderabad moving into GenAI and agentic AI roles, it is an early way to show you understand how agents connect to real systems.
This guide follows the official MCPA exam page. It covers the exam facts, the five domains and their weightings, the core MCP concepts with JSON-RPC examples, the security points, a 4-week plan and 15 original practice questions.
Quick facts (Linux Foundation, November 2026)
| Item | Detail |
|---|---|
| Format | Online, proctored, multiple choice |
| Duration | 90 minutes |
| Price | US$250 for the exam (US$495 with a THRIVE-ONE subscription) |
| Specification | MCP 2026-07-28 |
| Included | One retake; 12 months to take the exam |
| Prerequisites | None (JSON-RPC, LLM APIs, agent concepts and OAuth basics help) |
| Validity | 2 years |
Read the official MCPA exam page and the MCP specification. Study the 2026-07-28 release, because the exam follows that version. This guide covers the core concepts that all recent releases share.
Domains and weightings
| Domain | Weight | Competencies |
|---|---|---|
| Interactions and Execution | 26% | Interaction patterns and response handling; error handling; the tool invocation lifecycle; protocol primitives |
| Security and Governance | 24% | Trust boundaries; permissions and consent; risk and safety controls; auditability and observability |
| Use Cases and Ecosystem | 20% | Roles, responsibilities and adoption; operational use cases; ecosystem and portability |
| MCP Fundamentals | 16% | Purpose and scope; core concepts; interoperability and value |
| Architecture and Components | 14% | Schemas and structured data; hosts, clients and servers; the model interaction flow |
Core MCP concepts
| Concept | What it means |
|---|---|
| Host | The AI application the user works in, such as an IDE assistant or a chat app. It manages connections and user consent. |
| Client | A connection inside the host, one per server, that speaks MCP to that server. |
| Server | A program that exposes capabilities, such as a GitHub, database or file-system server. |
| Server primitives | Tools (actions the model can call), resources (data the application can read) and prompts (reusable templates the user can pick) |
| Client primitives | Features a server can ask the client for, such as sampling (asking the host's model for a completion), roots (which folders or URIs the server may use) and elicitation (asking the user for input) |
| Messages | JSON-RPC 2.0 requests, responses and notifications |
| Lifecycle | An initialize request with capability negotiation, an initialized notification, normal operation, then shutdown |
| Transports | stdio for local servers; Streamable HTTP for remote servers |
A tool call and its result:
// client → server
{"jsonrpc": "2.0", "id": 7, "method": "tools/call",
"params": {"name": "search_jobs", "arguments": {"city": "Hyderabad", "role": "DevOps"}}}
// server → client
{"jsonrpc": "2.0", "id": 7,
"result": {"content": [{"type": "text", "text": "12 open DevOps roles found"}],
"isError": false}}
Errors come in two kinds. A protocol problem, such as an unknown method or invalid parameters, returns a JSON-RPC error object. A tool that ran but failed, such as an API returning a timeout, returns a normal result with isError: true, so the model can see what went wrong and try something else.
How a tool is described to the client (from tools/list):
{"name": "search_jobs",
"description": "Search open job listings by city and role",
"inputSchema": {"type": "object",
"properties": {"city": {"type": "string"}, "role": {"type": "string"}},
"required": ["city"]}}
Security points the exam stresses
- Trust boundaries: treat tool descriptions and tool results from a server as untrusted input. They can carry prompt injection.
- Consent: the host should show users what a tool will do and get approval for sensitive actions. Servers should only get the roots and data they need.
- Authorisation for remote servers: HTTP transports use OAuth 2.1. Tokens should be issued for that specific server and never passed through to other APIs.
- Least privilege: give each server narrow scopes and read-only access where possible. Run local servers with limited permissions.
- Auditability: log tool calls, arguments, results and approvals so you can trace what an agent did and why.
4-week study plan
| Week | Focus | Hands-on |
|---|---|---|
| 1 | Why MCP exists; hosts, clients and servers; JSON-RPC 2.0; lifecycle and capability negotiation | Connect an existing MCP server to an MCP-capable host and read its logs |
| 2 | Tools, resources and prompts; sampling, roots and elicitation; schemas and structured output | Build a small server with the official Python or TypeScript SDK that exposes one tool and one resource |
| 3 | Tool invocation lifecycle, error handling, notifications, progress and cancellation; stdio vs Streamable HTTP | Inspect messages with the MCP Inspector; make the tool fail and check how errors return |
| 4 | Trust boundaries, consent, OAuth 2.1, prompt injection, auditing; use cases and ecosystem; review | Write a threat model for your server; book the exam |
15 practice questions with answers
These are original questions written for this guide, not exam questions.
1. What problem does MCP solve?
It gives AI applications one standard way to connect to tools and data, instead of a custom integration for every pair of app and system.
2. Which component does the user interact with directly?
The host.
3. How many servers does one MCP client connect to?
One. A host creates a separate client for each server.
4. Which message format does MCP use?
JSON-RPC 2.0.
5. What happens during initialisation?
Client and server agree on a protocol version and declare their capabilities, then the client sends an initialized notification.
6. Which primitive lets the model take an action, such as creating a ticket?
A tool.
7. Which primitive exposes read-only data, such as a file or a database schema?
A resource.
8. Which primitive is a user-selected template?
A prompt.
9. Which client feature lets a server request an LLM completion through the host?
Sampling.
10. Which client feature lets a server ask the user for missing information?
Elicitation.
11. Which transport suits a local server started by the host?
stdio.
12. A tool's external API times out. How should the server report it?
As a tool result with isError: true and a helpful message, not as a JSON-RPC protocol error.
13. How does a client learn a tool's arguments?
From the tool's inputSchema, a JSON Schema returned by tools/list.
14. Why treat tool results as untrusted?
They may contain injected instructions that try to make the model misuse other tools or leak data.
15. Which authorisation framework does MCP use for HTTP-based servers?
OAuth 2.1.
Before and after MCPA
- Learn GenAI foundations with Google Cloud Generative AI Leader or AWS AI Practitioner.
- Build agents professionally with AI-103, AWS Generative AI Developer and Databricks Generative AI Engineer.
- Learn it hands-on in our AI, GenAI and Agentic AI course and Python course; see live AI and GenAI jobs.
Cloud Soft Solutions in Ameerpet, Hyderabad teaches Python, GenAI, agentic AI and cloud with real-time projects. See the APEX program and all courses.
📞 Call +91 96660 19191 · WhatsApp · Flat No. 513, 5th Floor, Nilagiri Block, Aditya Enclave, Ameerpet, Hyderabad
Frequently Asked Questions
What is MCPA?
The Model Context Protocol Associate, the first certification for MCP, launched by the Agentic AI Foundation with Linux Foundation Training and Certification.
How long is the MCPA exam and what does it cost?
90 minutes of multiple-choice questions, US$250, with one retake.
Which MCP version does MCPA follow?
The MCP specification release 2026-07-28.
What are the MCPA domains?
Interactions and Execution 26%, Security and Governance 24%, Use Cases and Ecosystem 20%, MCP Fundamentals 16%, Architecture and Components 14%.
Do I need coding experience for MCPA?
No prerequisites are required. Knowing JSON-RPC, LLM APIs, agent concepts and OAuth basics helps.
How long is MCPA valid?
2 years.



