Unlocking AI Agent Productivity: Custom MCP Server Integration with CRM

How an MCP Server Solves AI Agent Integration Problems

Imagine 50 sales managers spending an average of 15 minutes gathering customer information before a call. That's 500 hours of unproductive work per month. They switch between Claude and CRM, manually copy data, and lose focus. An MCP server solves this — it gives the AI assistant direct access to corporate data via a single protocol. Preparation time is reduced by 73% (based on internal benchmarks), resulting in monthly savings of approximately $15,000 for a 50-person team.

We develop custom MCP intermediaries turnkey. We handle the full cycle: from design to deployment and documentation. We guarantee security and compatibility with any MCP client (Claude, Cursor, custom agents). Experience: over five years in AI integrations and 50+ successful projects.

An MCP server acts as a bridge: you describe tools (functions), resources (data), and prompts (templates), and the agent calls them as ordinary methods. One server — implement once, works with all MCP clients.

Why HTTP+SSE Transport Is Preferred for Production

For production, we always choose HTTP+SSE with authorization — this guarantees that data does not leak outside the corporate network. Meanwhile, stdio transport is 2–3 times faster in latency but does not scale. HTTP+SSE is 2x more secure due to built-in authentication.

Example MCP Server in Python with Two Transports

Below is a minimal MCP server in Python that provides tools for working with CRM. The mcp library (FastMCP) allows defining tools, resources, and prompts in a few lines.

# pip install mcp
from mcp.server.fastmcp import FastMCP
from mcp import types
import json

mcp = FastMCP("Corporate CRM Server")

@mcp.tool()
async def search_customers(
    query: str,
    limit: int = 10,
    status: str = "active",
) -> list[dict]:
    """Search customers in CRM by name or company."""
    results = await crm_db.search(
        query=query,
        limit=min(limit, 50),
        status=status,
    )
    return [{"id": r.id, "name": r.name, "company": r.company, "email": r.email}
            for r in results]

@mcp.tool()
async def get_customer_orders(customer_id: str, months: int = 3) -> dict:
    """Customer order history."""
    orders = await orders_db.get_for_customer(customer_id, months=months)
    return {
        "customer_id": customer_id,
        "total_orders": len(orders),
        "total_revenue": sum(o.amount for o in orders),
        "orders": [{"id": o.id, "date": str(o.date), "amount": o.amount, "status": o.status}
                   for o in orders[:20]],
    }

@mcp.tool()
async def create_task(
    customer_id: str, title: str, description: str, assignee: str, due_date: str,
) -> dict:
    """Create a task in CRM."""
    task = await crm_tasks.create(...)
    return {"task_id": task.id, "url": task.url}

@mcp.resource("crm://dashboards/{dashboard_id}")
async def get_dashboard(dashboard_id: str) -> str:
    data = await crm_analytics.get_dashboard(dashboard_id)
    return json.dumps(data, ensure_ascii=False, indent=2)

@mcp.resource("crm://segments")
async def list_segments() -> str:
    segments = await crm_db.get_segments()
    return json.dumps([{"id": s.id, "name": s.name, "count": s.count}], ensure_ascii=False)

@mcp.prompt()
def customer_analysis_prompt(customer_id: str) -> list[types.PromptMessage]:
    return [
        types.PromptMessage(
            role="user",
            content=types.TextContent(
                type="text",
                text=f"Perform a full analysis of customer {customer_id}:\n1. Order history and trends\n2. Churn risk\n3. Upsell\n4. Recommendations"
            ),
        )
    ]

if __name__ == "__main__":
    mcp.run()

from mcp.server.fastmcp import FastMCP
from fastapi import FastAPI
import uvicorn

mcp = FastMCP("Remote CRM Server")
# ... tools, resources ...
app = mcp.get_asgi_app()

from fastapi.middleware.base import BaseHTTPMiddleware

class AuthMiddleware(BaseHTTPMiddleware):
    async def dispatch(self, request, call_next):
        token = request.headers.get("Authorization", "").replace("Bearer ", "")
        if not await verify_token(token):
            return Response(status_code=401)
        return await call_next(request)

app.add_middleware(AuthMiddleware)

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8080)

Case Study: Integrating an MCP Server with Bitrix24 for a Sales Department

From our practice: a company with 50 sales managers. They actively use Claude to prepare for meetings, but each time they manually search for information in Bitrix24. This took an average of 15 minutes per call.

We deployed a single MCP server on the corporate server, connected it to the Bitrix24 REST API. The server provides tools: customer search by name, deal history, task creation, segment analytics. Each manager connects via Claude Desktop — after authorization, the AI agent itself gets the needed data.

Results:

• Call preparation time reduced from 15 to 4 minutes (73% savings, 3x faster)
• System switching reduced by 83%
• Implementation required no training — Claude understands natural questions

Development Process and Timelines

Follow these steps to implement a custom MCP server:

1. Analyze requirements and define tools/resources (1–2 days)
2. Develop MCP server with chosen transport (2–5 days)
3. Integrate with customer system (1–3 days)
4. Test, deploy, and document (1–2 days)

A basic server on stdio — from 1 day. Full integration with HTTP+SSE and authorization — from 5 days. Exact timelines are discussed after analyzing your API.

What's Included in the Deliverables

• Architecture documentation with diagrams and API specs
• Source code with unit tests and integration tests
• Deployment guide and operational runbook with monitoring setup
• Access credentials and security configuration (JWT, OAuth2, etc.)
• Team training session (1–2 hours) on using the MCP server with Claude Desktop
• 2 weeks of post-deployment support with SLA for critical issues

Security Risks We Address

Security risks include prompt injection and data leaks. We mitigate them with JWT authorization at the level of each tool, validation of input parameters, and rate limiting to prevent abuse. Additionally, HTTPS and resource isolation are configured. For high-load systems, we conduct load testing with p99 latency control — server response time should not exceed 200 ms.

We have been doing AI integrations for over five years, completed 50+ projects for companies from startups to enterprise. We use a proven stack: Python/TypeScript, authorization via JWT, ready integration with MCP SDK. We guarantee security and compatibility. Contact us to evaluate your project — we will prepare a commercial proposal within 1 business day. Order a custom MCP server development and save hours for your managers.

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.