Eonevolve Learning
Unity and DevelopmentUnity and codeBeginner~6 min

MCP Tools, Resources, and Prompts

Snapshot

~75 sec

MCP servers can expose tools, resources, and prompts. Tools request actions, resources provide context, and prompts offer reusable message templates. A client first discovers what a server offers, then selects the smallest primitive that matches the developer's intent.

You will learn

  • Distinguish an action-oriented tool from read-oriented resource context and a reusable prompt template.
  • Trace discovery before primitive use instead of assuming a fixed server surface.
  • Choose an appropriate primitive for a console read, project summary, and diagnosis workflow.

Target outcome

You can choose tools, resources, and prompts by responsibility rather than treating every MCP capability as a tool call.

Visual walkthrough

Visual walkthrough

~6 min

  1. Three primitives, three jobs

    • Tool

      A callable operation with declared inputs and outputs, such as reading filtered console entries.

    • Resource

      Addressable context a client can read, such as a read-only summary of project version and active build target.

    • Prompt

      A reusable message template that helps structure a workflow, such as a generic scene-diagnosis checklist.

  2. Discover first, then use the smallest primitive

    The client learns the connected server's current surface before it selects a primitive. Discovery is not permission to use every capability.

    1. Name the intent

      Example: explain the newest error without changing the project.

    2. Discover capabilities

      The client requests the available tools, resources, or prompts from the server.

    3. Choose one primitive

      Select read_console rather than loading unrelated context or invoking a write.

      read_console(severity: "error", limit: 10)

    4. Inspect the result

      Use the bounded output as evidence, then verify the diagnosis before acting.

    The client learns the connected server's current surface before it selects a primitive. Discovery is not permission to use every capability.

    Reading order

    1. Name the intent

      Example: explain the newest error without changing the project.

    2. Discover capabilities

      The client requests the available tools, resources, or prompts from the server.

    3. Choose one primitive

      Select read_console rather than loading unrelated context or invoking a write.

      read_console(severity: "error", limit: 10)
    4. Inspect the result

      Use the bounded output as evidence, then verify the diagnosis before acting.

    Connection explanations

    1. Name the intent → Discover capabilities (scopes)

      A clear intent keeps discovery focused on the capability actually needed.

    2. Discover capabilities → Choose one primitive (offers)

      Discovery tells the client what exists; the client still selects only the matching primitive.

    3. Choose one primitive → Inspect the result (returns)

      A primitive returns context or a result that must be interpreted and verified.

  4. Choose by responsibility

    ApproachPrimary jobGame-development exampleControl question
    ToolRequest a bounded operation from the server.Perform an action or query with explicit arguments and a structured result.read_console filters an invented Unity console by severity and result limit.Is the operation read-only or a write, and what approval or validation does it require?
    ResourceRead addressable context without pretending it is an action.Provide context the client can identify and retrieve when it becomes relevant.project://summary returns a read-only engine version, active target, and package-count summary.Is the resource bounded, current, and free of secrets or unrelated project data?
    PromptOffer a reusable structure for a human-guided interaction.Prepare messages or questions for a repeatable workflow without performing the work itself.diagnose_scene_issue asks for the symptom, relevant object, evidence, and a verification plan.Does the template clarify the task without implying that its recommendations are already approved?

  6. Choose the right primitive

    A client needs three capabilities: fetch the latest filtered console errors, read a stable project summary, and offer a repeatable diagnosis questionnaire. Which primitive fits each one?

    Expected reasoning

    Use a tool for the bounded read_console query, a resource for the read-only project summary, and a prompt for the reusable diagnosis questionnaire. The server may expose all three, but discovery and host policy still determine what the client can use.

Go deeper