Tool Integration in MCP

Beyond Simple Tools

MCP servers provide more than just tools:

1. Tools - Function execution
2. Resources - Data/files
3. Prompts - Pre-written instructions
4. Sampling - LLM inference

Resources

Expose data without exposing internals:

@server.resource("file")
def read_file(uri: str):
    # uri format: "file:///path/to/file"
    path = uri.replace("file://", "")
    with open(path, 'r') as f:
        return f.read()

@server.resource("document")
def get_document(doc_id: str):
    return db.documents.get(doc_id)

Client can access without knowing implementation!

Prompts

Pre-structured prompts for common tasks:

@server.prompt("summarize")
def summarize_prompt(text: str, style: str = "concise"):
    return f"""
    Summarize the following text in {style} style:
    
    {text}
    
    Summary:
    """

@server.prompt("test_generator")
def generate_tests_prompt(code: str):
    return f"""
    Generate unit tests for this code:
    
    {code}
    
    Tests:
    """

Tool Chaining

MCP servers can call other tools:

@server.tool("analyze_data")
def analyze(file_path: str):
    # This tool uses other tools
    data = fetch_file(file_path)  # Tool 1
    stats = calculate_stats(data)  # Tool 2
    report = format_report(stats)  # Tool 3
    return report

Streaming Results

For long-running operations:

@server.tool("long_operation", streaming=True)
def process_large_dataset(dataset: str):
    for chunk in chunks(dataset):
        result = process(chunk)
        yield result  # Send partial results
    
# Client receives results as they arrive!

Batch Operations

Execute multiple tool calls efficiently:

@server.tool("batch_search")
def search_multiple(queries: list[str]):
    results = []
    for query in queries:
        results.append(search(query))
    return results

# More efficient than calling search() multiple times

Tool Combinations

Server can combine tools for complex tasks:

@server.tool("research")
def research_topic(topic: str):
    # Combine multiple tools
    papers = search_arxiv(topic)
    summaries = [summarize(p) for p in papers]
    synthesis = synthesize_findings(summaries)
    return synthesis