presentsUNAIVERSE
The UN AI timeline

    Learn AI /AI Agents and Automation /Model Context Protocol

    Model Context ProtocolMCP

    Also called: MCP.

    Detail

    In five seconds

    What it is

    A standard way for an AI application to connect with external tools and data sources through a shared, predictable interface.

    In plain words

    MCP is a common way to let an AI application work with outside tools and information sources without building a completely different connection for each one.

    An everyday example

    At a food distribution point, a coordinator needs current stock levels, truck arrival times, and a map of nearby roads. Instead of calling three different offices with three different forms, the coordinator uses one service desk with a standard request sheet. The desk tells the coordinator which storerooms and services are available, what each can provide, and the correct format for each request. The replies come back in a predictable structure, so the coordinator can combine them into one plan for the day.

    Where the story stops working

    A service desk suggests a human clerk who understands the request and makes judgement calls. MCP does not work like that. It is a technical protocol with defined message formats and capabilities. Also, real MCP setups can support ongoing interaction between client and server, not just one paper form in and one reply out.

    Remember thisMCP standardises how an AI application discovers and invokes external resources and tools. It can simplify integration, but it does not itself provide approval, access control or data protection.

    This term is contested

    The broad idea of MCP is clear, but terminology and scope are still evolving in practice. Different people may use the term narrowly for a specific protocol implementation or more broadly for the surrounding ecosystem of AI-tool integration patterns.

    Understand it

    Understand it

    Why it matters

    For non-specialists, MCP matters because it affects cost, speed, control, and risk. If an organisation wants one AI application to work with many internal and external systems, a shared protocol can reduce one-off integration work and make changes easier to manage. It can also improve transparency by making tool access more structured and easier to govern. At the same time, MCP does not solve trust and safety by itself. Leaders still need to ask who can access which systems, what gets logged, when a human must approve an action, and what happens if the connected data is wrong or out of date. In decision-making terms, MCP is about whether your AI setup can connect to the right tools in a manageable way, not whether the output should be accepted without checking.

    How MCP links an AI application to outside tools

    After viewing this, the reader should be able to say how an AI application uses MCP to discover tools, request context, and return a result.

    1. 01

      1. User request

      A person asks for something that may require outside information or an action, such as checking a file or looking up a timetable.

      Place a box on the far left labelled 'User request'. Add a short example beneath it. Draw an arrow to the next box.

    2. 02

      2. AI application interprets the need

      The application identifies that it may need external context or a tool, rather than only generating text from the prompt.

      Add a second box labelled 'AI application'. Include a small note: 'Needs outside data or tool'. Draw an arrow to the next box.

    3. 03

      3. MCP client asks what is available

      The client side of the protocol contacts an MCP-compatible server to learn which tools or resources exist and how to call them.

      Add a third box labelled 'MCP client'. Draw an arrow to a fourth box. Add a small side label on the arrow: 'discover capabilities'.

    4. 04

      4. MCP server exposes tools and resources

      The server presents a standard description of available resources, such as files, databases, or actions, and returns structured results when called.

      Add a fourth box labelled 'MCP server'. Beneath it, place three small icons or mini-boxes labelled 'files', 'database', and 'calendar'. Draw arrows from these into the MCP server box.

    5. 05

      5. Result returns for answer or review

      The application combines the returned context or tool output into an answer, summary, or action proposal, often with human review where needed.

      Draw arrows back from 'MCP server' to 'MCP client' to 'AI application'. End with a right-side output box labelled 'Answer / draft / proposed action' and a small human-check icon beside it.

    A left-to-right flow diagram with five boxes: user request, AI application, MCP client, MCP server, and external tools/data. Arrows show the request moving right for tool discovery and use, then results moving left into the final answer or proposed action.

    On hover or tap, each box reveals one sentence defining its role. Clicking the MCP server highlights the connected tools to show that one protocol can expose many resources.

    Describe this without the diagram

    The diagram is a five-step flow read from left to right. First, a user makes a request. Second, the AI application recognises it may need outside information or a tool. Third, an MCP client contacts an MCP server to discover what tools and resources are available and how to use them. Fourth, the MCP server exposes connected systems such as files, databases, and calendars through a standard interface and returns structured results. Fifth, those results go back to the AI application, which produces an answer, a draft, or a proposed action, ideally with human review for important tasks.

    With reduced motion: A static left-to-right flowchart showing the same five boxes and arrows, with all labels visible at once and no animation or hover effects.

    MCP is a technical standard for communication between an AI application's MCP client and MCP servers. Servers can expose tools, resources, and prompts in a standard form. They may connect to files, databases, calendars, maps, or business software. Back-end systems still need server-side integration, but clients can use common protocol rules.

    A user request may lead the application to list available tools or resources through MCP. It can then read a resource or invoke a tool using its declared inputs. The user request itself is not necessarily an MCP input. MCP standardises these exchanges. It does not determine whether a call is safe.

    The application may use returned content or tool results to produce an answer or proposed action. MCP does not guarantee correctness and does not replace access control, logging, or human oversight.

    MCP is not simply an API, connector, or function calling. An API lets systems exchange data or actions. A connector is a ready-made link to a service. Function calling lets a model produce a structured tool request. MCP standardises how clients and servers discover and use exposed capabilities.

    How “A shared service desk for many storerooms” maps onto the real thing

    In the storyIn the AI system
    the service desk with one standard request sheetMCP as a shared protocol for accessing many tools and data sources
    storerooms, transport office, and map officedifferent external systems such as files, calendars, databases, or mapping services
    the desk listing what each office can providetool and resource discovery through a standard interface
    predictable reply formsstructured responses that an AI application can process consistently

    A worked example

    A protection officer evaluating an assistant with the procurement unit asks it to assemble the approved, aggregate inputs for a situation report before the bulletin file closes.

    Starting from: User request: 'Use the cleared template and the latest aggregate activity data to prepare the situation-report sections. Flag missing submissions rather than filling the gaps.'

    1. The application's MCP client connects only to the servers approved for the evaluation.
    2. The servers advertise their available capabilities: one exposes the cleared template as a resource, while another exposes a tool for querying aggregate activity data.
    3. The client invokes the data tool with the reporting period and agreed aggregation level. It receives a structured response containing activity totals and status fields for missing submissions.
    4. The client retrieves the template resource and maps the returned fields to the corresponding sections.
    5. The application drafts the sections, identifies gaps and presents the result to the protection officer for verification and clearance.

    Result: The draft follows the cleared structure, carries source labels and retrieval times, and marks missing inputs without inventing them. The officer still checks the figures, context and wording before the situation report enters clearance.

    In a United Nations workplace

    During a procurement unit's technical evaluation, a protection officer tests whether a proposed AI assistant can retrieve a cleared situation-report template and non-sensitive, aggregate activity figures through MCP before submissions for the afternoon regional bulletin are locked.

    MCP gives the assistant a predictable way to discover the available resource and tool interfaces, invoke them and receive structured results. The procurement unit can assess these connectors rather than accepting a separate bespoke integration for every dashboard or repository.

    Before you try thisMCP standardises the connection. It does not approve the system, grant access or make sensitive data safe. Case records, names, contact details, free-text accounts, survivor or witness information, and risk-revealing locations must not be exposed unless every component is expressly approved for that material. For the test, use only aggregate, non-sensitive data with least-privilege access, logging, retention controls and human review.

    Where you may hear it

    • A supplier demonstration during a procurement unit's technical evaluation
    • An ICT governance board reviewing connections between an assistant and internal systems
    • A Headquarters architecture review before an approved pilot
    • A developer call when replacing several custom connectors with one standard interface

    Avoid confusion

    Avoid confusion

    Often confused with

    Common misconceptions

    • MCP is just another name for an API.

      They are related but not identical. An API is usually the interface for one service or product. MCP is a standardised protocol intended to let AI applications discover and use many tools and resources in a more consistent way. A system may expose capabilities through MCP while those capabilities themselves still rely on underlying APIs.

    • If a system uses MCP, it is automatically safe and reliable.

      A protocol can standardise communication, but it does not guarantee correct outputs, secure configuration, or appropriate use. Safety still depends on access controls, approval steps, monitoring, data quality, and careful choice of which tools the AI application may use.

    • MCP means the AI model can directly do anything on connected systems.

      In practice, what it can access depends on what the MCP server exposes and what permissions are granted. The protocol can describe tools and resources, but organisations still decide which actions are allowed and whether human review is required.

    Quick check

    Which description best distinguishes MCP from a single connector or API?

    Go deeper

    Go deeper

    The simple version

    MCP is a common way to let an AI application work with outside tools and information sources without building a completely different connection for each one.

    The technical version

    Technically, MCP is a client-server protocol for exposing context, resources, prompts, and tool interfaces to AI applications in a standard form. An MCP client, often inside an AI application, connects to an MCP server, which advertises available capabilities and returns structured responses when those capabilities are invoked. This supports capability discovery and more uniform tool integration across different back-end systems. It is commonly discussed alongside tool use and function calling, but it operates at a broader integration layer: function calling is often the model-level mechanism for selecting a tool, while MCP is the protocol that can describe and deliver the tool definitions and resource access in a consistent way. Exact implementations can vary, so the practical behaviour depends on the host application, permissions, and server design.

    Reviewed external resources

    • Primary sourceModel Context ProtocolModel Context Protocol

      This link is the official source and suits technical readers who want the primary specification and implementation-oriented details.

      Deeperlink checked 2026-07-31

    • EncyclopaediaModel Context ProtocolWikipedia

      This link gives a general overview of Model Context Protocol and suits readers who want a quick background summary before going deeper.

      Intermediatelink checked 2026-07-31

    • ArticleBuilding effective agentsAnthropic

      This link discusses agents and is useful for readers who want to place MCP in the broader context of tool-using AI systems.

      Intermediatelink checked 2026-07-31

    • VideoVideo explanations: Model Context ProtocolYouTube search

      This link is a YouTube search results page, so it helps readers find different video explanations and compare teaching styles rather than relying on one source.

      search results, not one pagelink checked 2026-07-31

    • DiscussionCommunity discussion: Model Context ProtocolReddit search

      This link is a Reddit search results page, so it suits readers who want to see community questions, examples, and debates, with the usual caution that discussion posts vary in quality.

      search results, not one pagelink checked 2026-07-31

    Where this came from

    Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.

    The one thing to remember

    MCP standardises how an AI application discovers and invokes external resources and tools. It can simplify integration, but it does not itself provide approval, access control or data protection.