Anthropic Claude Agent SDK Integration for AI Agents

Integration of Anthropic Claude Agent SDK for Building AI Agents

Let's be direct: when a client comes with a task to build an agent that doesn't just answer questions but performs actions—searching databases, creating tickets, working with files—we know that the Anthropic Claude Agent SDK removes 80% of the grunt work. But incorrect integration leads to context loss and hallucinations. Here's how we configure the SDK to make the agent reliable in production. Our team has 5+ years of experience in NLP and LLMs, with over 20 projects using the Claude Agent SDK. We guarantee stable agent operation in your infrastructure.

Problems We Solve

• Context drift in multi-turn dialogs: Without proper session management, the agent forgets previous interactions, causing repetitive questions or incorrect actions. The SDK's built-in ConversationSession handles this automatically.
• Tool integration complexity: Manually wrapping each API call with error handling, retries, and parsing is error-prone and time-consuming. The @tool decorator streamlines this.
• Lack of human oversight for critical operations: Destructive actions like deleting orders or processing refunds need approval. The SDK's ToolApprovalPolicy provides a clean interface for human-in-the-loop.

How We Do It: Technical Details

We start by installing the SDK and defining tools via decorators. Below is a typical template we use.

# pip install anthropic claude-agent-sdk
import anthropic
from claude_agent_sdk import Agent, AgentConfig, tool

client = anthropic.Anthropic()

@tool
def search_database(query: str, table: str = "products") -> str:
    """Search the company database.

    Args:
        query: Search query
        table: Table to search (products, orders, customers)
    """
    results = db.search(query=query, table=table, limit=10)
    return results.to_json()

@tool
def create_support_ticket(
    customer_id: str,
    subject: str,
    description: str,
    priority: str = "normal",
) -> str:
    """Create a support ticket.

    Args:
        customer_id: Customer ID
        subject: Ticket subject
        description: Detailed description
        priority: Priority (low, normal, high, critical)
    """
    ticket = helpdesk.create_ticket(
        customer_id=customer_id,
        subject=subject,
        description=description,
        priority=priority,
    )
    return f"Ticket #{ticket['id']} created. URL: {ticket['url']}"

config = AgentConfig(
    model="claude-opus-4-5",
    system_prompt="""You are a customer support agent for TechCorp.
Help customers solve problems using available tools.
Always verify data through tools—do not rely on memory.""",
    max_turns=10,
)

agent = Agent(
    client=client,
    config=config,
    tools=[search_database, create_support_ticket],
)

result = agent.run(
    messages=[{"role": "user", "content": "Customer ID 12345 has a problem with order #99876"}]
)
print(result.final_message)

For real-time interactions, we use streaming via astream.

import asyncio

async def run_agent_with_streaming():
    async for event in agent.astream(
        messages=[{"role": "user", "content": "Analyze the last 10 orders for customer ID 12345"}]
    ):
        match event.type:
            case "text_delta":
                print(event.text, end="", flush=True)
            case "tool_use_start":
                print(f"\n[Tool: {event.tool_name}]")
            case "tool_result":
                print(f"[Result received, {len(event.content)} chars]")
            case "agent_turn_complete":
                print(f"\n[Completed in {event.turn_count} turns]")

asyncio.run(run_agent_with_streaming())

MCP Integration

Model Context Protocol (MCP) allows connecting external servers with tools without explicit coding. We use this for filesystem, database, and GitHub integration.

from claude_agent_sdk import Agent, MCPServerConfig

agent_with_mcp = Agent(
    client=client,
    config=config,
    mcp_servers=[
        MCPServerConfig(
            name="filesystem",
            command="npx",
            args=["-y", "@modelcontextprotocol/server-filesystem", "/workspace"],
        ),
        MCPServerConfig(
            name="postgres",
            command="npx",
            args=["-y", "@modelcontextprotocol/server-postgres"],
            env={"POSTGRES_URL": "postgresql://user:pass@localhost/db"},
        ),
        MCPServerConfig(
            name="github",
            command="npx",
            args=["-y", "@modelcontextprotocol/server-github"],
            env={"GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."},
        ),
    ],
)

result = agent_with_mcp.run(
    messages=[{"role": "user", "content": "Read the file config.yaml and create GitHub Issues from TODO comments"}]
)

Multi-Turn Dialog with History

We use ConversationSession to maintain context across turns without manual history management.

from claude_agent_sdk import ConversationSession

session = ConversationSession(
    agent=agent,
    session_id="customer_session_12345",
)

response1 = session.send("What is the status of my order #99876?")
response2 = session.send("Can I reschedule delivery for tomorrow?")
response3 = session.send("Please confirm")

print(session.get_history())

Human-in-the-Loop via Approval

For destructive operations, we require approval. The ToolApprovalPolicy lets us define which tools need confirmation.

from claude_agent_sdk import Agent, ToolApprovalPolicy

class CustomApprovalPolicy(ToolApprovalPolicy):
    REQUIRES_APPROVAL = {"delete_order", "process_refund", "ban_customer"}

    async def should_approve(self, tool_name: str, tool_input: dict) -> bool:
        if tool_name not in self.REQUIRES_APPROVAL:
            return True
        await notify_operator(
            message=f"Approval required: {tool_name}\nParameters: {tool_input}",
            callback_url="/api/approve/{approval_id}",
        )
        approval = await wait_for_approval(timeout=300)
        return approval.approved

agent_with_approval = Agent(
    client=client,
    config=config,
    tools=[search_database, process_refund, ban_customer],
    approval_policy=CustomApprovalPolicy(),
)

Practical Case: Financial Monitoring Agent

Comparison: Manual Implementation vs. Claude Agent SDK

Using the SDK cuts agent development time by three times compared to manual implementation. Based on our data, clients achieve substantial annual savings on manual processing.

Common Mistakes and Solutions

Process and Timeline

• Architectural design of the agent for your scenario — 1–2 days.
• SDK integration with your infrastructure and tool setup — 3–5 days.
• Connecting MCP servers (files, databases, external APIs) — 1–3 days each.
• Setting up human-in-the-loop and approval flow — 1 week.
• Documentation and team training — 2–3 days.
• Production deployment with monitoring — 1 week.
• Support for 1 month after deployment is included.

Total timeline: 2 to 4 weeks depending on complexity. Exact cost is calculated individually after analyzing your scenario.

What You Get

• A working agent with configured tools and MCP servers.
• Full documentation on architecture and API.
• Access to source code and CI/CD pipeline.
• Team training (2–3 sessions).
• One month of technical support after deployment.

Why Integrate the Claude Agent SDK?

The SDK provides out-of-the-box mechanisms for context management, tools, and security, accelerating production release by three times. According to an internal client survey, ticket processing time decreases by 60%. Contact us to assess your scenario and schedule a consultation for integration.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.