Composition + Sampling Lab

Explore how MCP tools, resources, and prompts can be composed into a workflow and how the MCP client can provide Sampling.

GuidedNo API keys in Lab

What happens when MCP results are combined and then used to request Sampling?

What do you want to see?

Read a resource, pass it into composition, then request Sampling.

Deterministic replay · No network

Protocol trace11 events

Recorded MCP exchange — observable protocol events only. No chain-of-thought.

1. initialize

{
  "interactionId": "initialize",
  "method": "initialize",
  "clientInfo": {
    "name": "dataaihub-cookbook-mcp-client",
    "version": "0.1.0"
  },
  "samplingCallbackRegistered": true,
  "samplingMode": "mock"
}

2. initialize

{
  "interactionId": "initialize",
  "method": "initialize",
  "protocolVersion": "2025-11-25",
  "serverInfo": {
    "name": "dataaihub-cookbook-composition",
    "version": ""
  },
  "capabilities": {
    "experimental": {},
    "prompts": {
      "list_changed": false
    },
    "resources": {
      "subscribe": false,
      "list_changed": false
    },
    "tools": {
      "list_changed": false
    }
  }
}

3. resources/list

{
  "interactionId": "resources/list",
  "method": "resources/list"
}

4. resources/list

{
  "interactionId": "resources/list",
  "method": "resources/list",
  "resources": [
    {
      "uri": "acme://docs/knowledge-platform",
      "name": "knowledge-platform",
      "description": "Service documentation for the Knowledge Platform.",
      "mimeType": "text/markdown"
    },
    {
      "uri": "acme://docs/billing-portal",
      "name": "billing-portal",
      "description": "Service documentation for the Billing Portal.",
      "mimeType": "text/markdown"
    },
    {
      "uri": "acme://status/services",
      "name": "service-status",
      "description": "Service statuses for the Acme AI environment.",
      "mimeType": "application/json"
    }
  ],
  "resourceCount": 3
}

5. resources/read

{
  "interactionId": "resources/read:acme://docs/knowledge-platform",
  "method": "resources/read",
  "uri": "acme://docs/knowledge-platform"
}

6. resources/read

{
  "interactionId": "resources/read:acme://docs/knowledge-platform",
  "method": "resources/read",
  "uri": "acme://docs/knowledge-platform",
  "isError": false,
  "contents": [
    {
      "uri": "acme://docs/knowledge-platform",
      "text": "# Knowledge Platform\n\nService documentation for the Knowledge Platform.\n\nThe Knowledge Platform indexes internal documents and serves retrieval\nqueries for Acme applications. Clients read this resource for current\nservice facts; they do not invent the catalog locally.\n\n## Services\n\n- knowledge-platform\n- billing-api\n- identity-api\n\n## Capabilities\n\n- Document ingest and chunking\n- Keyword and vector retrieval\n- Citation-preserving answer assembly\n\n## Operational notes\n\nStatus and incident details live in `acme://status/services`.\nBilling product surfaces live in `acme://docs/billing-portal`.\n",
      "mimeType": "text/markdown"
    }
  ]
}

7. tools/call

{
  "interactionId": "tools/call:compose_resource_brief:2",
  "method": "tools/call",
  "name": "compose_resource_brief",
  "arguments": {
    "uri": "acme://docs/knowledge-platform",
    "content": "# Knowledge Platform\n\nService documentation for the Knowledge Platform.\n\nThe Knowledge Platform indexes internal documents and serves retrieval\nqueries for Acme applications. Clients read this resource for current\nservice facts; they do not invent the catalog locally.\n\n## Services\n\n- knowledge-platform\n- billing-api\n- identity-api\n\n## Capabilities\n\n- Document ingest and chunking\n- Keyword and vector retrieval\n- Citation-preserving answer assembly\n\n## Operational notes\n\nStatus and incident details live in `acme://status/services`.\nBilling product surfaces live in `acme://docs/billing-portal`.\n",
    "mime_type": "text/markdown",
    "name": "knowledge-platform"
  }
}

8. sampling/createMessage

{
  "interactionId": "sampling/createMessage:1",
  "method": "sampling/createMessage",
  "params": {
    "meta": {},
    "messages": [
      {
        "role": "user",
        "content": {
          "type": "text",
          "text": "Resource URI: acme://docs/knowledge-platform\nName: knowledge-platform\nMIME type: text/markdown\nContent:\n# Knowledge Platform\n\nService documentation for the Knowledge Platform.\n\nThe Knowledge Platform indexes internal documents and serves retrieval\nqueries for Acme applications. Clients read this resource for current\nservice facts; they do not invent the catalog locally.\n\n## Services\n\n- knowledge-platform\n- billing-api\n- identity-api\n\n## Capabilities\n\n- Document ingest and chunking\n- Keyword and vector retrieval\n- Citation-preserving answer assembly\n\n## Operational notes\n\nStatus and incident details live in `acme://status/services`.\nBilling product surfaces live in `acme://docs/billing-portal`.\n\n\nWrite a two-sentence operational brief using only this resource."
        }
      }
    ],
    "system_prompt": "Use only the supplied MCP resource content. Do not invent facts.",
    "max_tokens": 200
  },
  "boundary": "mcp-client-sampling-callback"
}

9. sampling/createMessage

{
  "interactionId": "sampling/createMessage:1",
  "method": "sampling/createMessage",
  "isError": false,
  "result": {
    "role": "assistant",
    "content": {
      "type": "text",
      "text": "[mock] Client sampling callback completed this request.\nGrounded context (1 message(s)):\nResource URI: acme://docs/knowledge-platform\nName: knowledge-platform\nMIME type: text/markdown\nContent:\n# Knowledge Platform\n\nService documentation for the Knowledge Platform.\n\nThe Knowledge Platform indexes internal documents and serves retrieval\nqueries for Acme applications. Clients read this resource for current\nservice facts; they do not invent the catalog locally.\n\n## Services\n\n- knowledge-platform\n- billing-api\n- identity-api\n\n## Capabilities\n\n- Document ingest and chunking\n- Keyword and vector retrieval\n- Citation-preserving answer assembly\n\n## Operational notes\n\nStatus and incident details live in `acme://status/services`.\nBilling product surfaces live in `acme://docs/billing-portal`.\n\n\nWrite a two-sentence operational brief using only this resource."
    },
    "model": "mock"
  },
  "boundary": "mcp-client-sampling-callback"
}

10. tools/call

{
  "interactionId": "tools/call:compose_resource_brief:2",
  "method": "tools/call",
  "name": "compose_resource_brief",
  "arguments": {
    "uri": "acme://docs/knowledge-platform",
    "content": "# Knowledge Platform\n\nService documentation for the Knowledge Platform.\n\nThe Knowledge Platform indexes internal documents and serves retrieval\nqueries for Acme applications. Clients read this resource for current\nservice facts; they do not invent the catalog locally.\n\n## Services\n\n- knowledge-platform\n- billing-api\n- identity-api\n\n## Capabilities\n\n- Document ingest and chunking\n- Keyword and vector retrieval\n- Citation-preserving answer assembly\n\n## Operational notes\n\nStatus and incident details live in `acme://status/services`.\nBilling product surfaces live in `acme://docs/billing-portal`.\n",
    "mime_type": "text/markdown",
    "name": "knowledge-platform"
  },
  "isError": false,
  "result": {
    "isError": false,
    "content": [
      {
        "type": "text",
        "text": "{\n  \"ok\": true,\n  \"workflow\": \"resource_to_sampling\",\n  \"resourceUri\": \"acme://docs/knowledge-platform\",\n  \"resourceName\": \"knowledge-platform\",\n  \"mimeType\": \"text/markdown\",\n  \"sampling\": {\n    \"model\": \"mock\",\n    \"role\": \"assistant\",\n    \"text\": \"[mock] Client sampling callback completed this request.\\nGrounded context (1 message(s)):\\nResource URI: acme://docs/knowledge-platform\\nName: knowledge-platform\\nMIME type: text/markdown\\nContent:\\n# Knowledge Platform\\n\\nService documentation for the Knowledge Platform.\\n\\nThe Knowledge Platform indexes internal documents and serves retrieval\\nqueries for Acme applications. Clients read this resource for current\\nservice facts; they do not invent the catalog locally.\\n\\n## Services\\n\\n- knowledge-platform\\n- billing-api\\n- identity-api\\n\\n## Capabilities\\n\\n- Document ingest and chunking\\n- Keyword and vector retrieval\\n- Citation-preserving answer assembly\\n\\n## Operational notes\\n\\nStatus and incident details live in `acme://status/services`.\\nBilling product surfaces live in `acme://docs/billing-portal`.\\n\\n\\nWrite a two-sentence operational brief using only this resource.\"\n  }\n}"
      }
    ],
    "structuredContent": {
      "ok": true,
      "workflow": "resource_to_sampling",
      "resourceUri": "acme://docs/knowledge-platform",
      "resourceName": "knowledge-platform",
      "mimeType": "text/markdown",
      "sampling": {
        "model": "mock",
        "role": "assistant",
        "text": "[mock] Client sampling callback completed this request.\nGrounded context (1 message(s)):\nResource URI: acme://docs/knowledge-platform\nName: knowledge-platform\nMIME type: text/markdown\nContent:\n# Knowledge Platform\n\nService documentation for the Knowledge Platform.\n\nThe Knowledge Platform indexes internal documents and serves retrieval\nqueries for Acme applications. Clients read this resource for current\nservice facts; they do not invent the catalog locally.\n\n## Services\n\n- knowledge-platform\n- billing-api\n- identity-api\n\n## Capabilities\n\n- Document ingest and chunking\n- Keyword and vector retrieval\n- Citation-preserving answer assembly\n\n## Operational notes\n\nStatus and incident details live in `acme://status/services`.\nBilling product surfaces live in `acme://docs/billing-portal`.\n\n\nWrite a two-sentence operational brief using only this resource."
      }
    }
  }
}

11. termination
```
{
  "reason": "session_closed"
}
```

How it works

The server requests Sampling through MCP. The client controls how the model interaction is handled.

Run MCP #4 locally

The Lab replays measured traces. Use the Cookbook to run the same cases with the official MCP Python SDK. The default Lab never requires OPENAI_API_KEY.

cd examples/mcp/04-composition
uv sync --extra dev
uv run python main.py --case resource-to-sampling --show-sequence

View MCP #4 source

Previous Lab

MCP #3 covers prompt discovery and retrieval. This Lab composes those primitives and requests Sampling.

← Prompt Discovery & Retrieval

Continue Learning

Guides that explain MCP as a client/server protocol for tools, resources, prompts, and Sampling.

Related Tools

Tools commonly used with MCP hosts and clients.

Related Rankings

See how this stack compares in DataAIHub rankings.

Best AI Agent Frameworks

Lab ID mcp-composition · guided / precomputed mock