Mcp: implementation and integration

10k+active public MCP servers in the ecosystem in the first year after release

97MSDK downloads per month (Python + TypeScript)

714 → 16kgrowth in the number of MCP servers from January 2025 to 2026

1200+Quantium employees work with data through MCP agents

Industry solutions

What you can build with MCP

E-commerce and retail Connect the agent to PIM, warehouse and orders via MCP servers Answers on availability, prices and order statuses from live data instead of manual exportsLearn more →Manufacturing Wrap MES, ERP and technical documentation into MCP servers for agents Engineers and operators get answers on procedures and line status without searching across systemsLearn more →Finance and banking Give agents governed access to transactions and compliance data via MCP Risk scoring, fraud detection and compliance monitoring in real timeLearn more →Logistics Connect TMS, tracking and document workflow as MCP tools The agent builds delivery statuses and produces documents across the end-to-end processLearn more →B2B distribution Connect CRM, price lists and a customer's order history through a single MCP layer The manager and agent prepare proposals and upsells on the full customer contextLearn more →IT and development Connect Git, trackers and CI/CD to the agent via MCP The agent finds context across code and tasks, speeding up development and onboardingLearn more →Professional services Deploy Sloy as an MCP server for corporate memory across projects and people Decisions and project history are available to agents as verifiable memory; knowledge is not lost during staff rotationLearn more →Real estate Wrap asset databases, CRM and documents into MCP servers The agent selects assets and prepares documents from up-to-date portfolio dataLearn more →

Capabilities

MCP capabilities

Sources: CRM, Git, BI, files, email, tasks, financeMCP servers (unified interface + OAuth 2.1)Sloy — an MCP server for corporate memoryMCP protocol (two-way)AI agent / client: Claude, ChatGPT, Gemini, CopilotActions in systems + verifiable answers

On the left, corporate data sources and tools. Each is wrapped in an MCP server with a unified interface and authorization. On the right, an AI agent (any client: Claude/ChatGPT/Gemini/Copilot) reaches the servers through the MCP protocol, reads data and performs actions. Sloy connects as an MCP server for corporate memory. The arrows are bidirectional: reading context and performing actions.

One protocol instead of N×M integrations

A data source is connected once and works with all agents — integration cost and time drop, and one-off connectors stop multiplying.

Transferable integrations

MCP servers outlive a change of LLM, team or contractor: nothing is rewritten from scratch when a project is handed over.

Access to up-to-date enterprise data

The agent answers from live data in CRM, Git, BI and files instead of a stale training corpus — fewer hallucinations and manual checks.

Managed access and audit

OAuth 2.1, centralized authorization and call logging give security teams control over what the agent can access.

Two-way tools, not read-only

The agent not only reads data but also performs actions (create a task, update a record) — automating end-to-end processes, not just prompts.

Compatible with any client

One MCP server serves Claude, ChatGPT, Gemini, Cursor and Copilot — the company isn't locked into a single vendor.

Dynamic tool loading

The agent picks the right tools on the fly instead of hardcoded integrations — new scenarios launch without rewriting code.

Integration with Sloy corporate memory

The company's work trail (chats, meetings, Git, finance) is served to agents as verifiable memory via MCP — decisions are made on company context.

Mature international standard

MCP was handed to the Agentic AI Foundation (Linux Foundation) and backed by OpenAI, Google, Microsoft, AWS — a bet on a standard, not on custom in-house code.

Approach

How we implement MCP

Minimal core modification

We don't fork or patch the MCP core. MCP stays on the standard, updatable version — we move business logic into separate microservices alongside it, so platform updates don't break your customizations.

International Standards, Not Homegrown Hacks

Where a mature international solution exists, we use it instead of inventing our own protocol or platform. Before writing code, we study how the problem is already solved in the industry.

Transferability

The solution is loosely coupled and documented: it can be handed over between teams and contractors without rewriting. You are not tied to us.

AI compatibility

MCP in the AI stack

Open standard under the Linux Foundation

MCP was contributed by Anthropic to the Agentic AI Foundation; founding members are Block, OpenAI, AWS, Google, Microsoft. It is not a vendor protocol but community-driven infrastructure.

Native support in major clients

Claude, ChatGPT, Gemini, Microsoft Copilot, Cursor and VS Code have first-class MCP integration — one server works with all of them.

Adopted by major platforms

OpenAI (Agents SDK, Responses API, March 2025), Google DeepMind (Gemini, April 2025), Microsoft Azure AI Agent Service, AWS, Cloudflare, Bloomberg, Snowflake, Salesforce.

MCP in RFPs as a baseline requirement

By the end of 2025, MCP support moved from a competitive edge to a mandatory line in enterprise procurement requirements — plan for compatibility in advance.

Mature security mechanisms

OAuth 2.1 and centralized governance cover authorization; meanwhile MCP servers require separate supply-chain control — KT.Team connects sources through managed servers.

News

What's new in MCP

All news

21.05.2026
MCP 2026-07-28 RC: stateless core, session ID and initialize handshake removed
The RC removes the initialize handshake and sessions: state is passed via explicit handles, follow-ups go through InputRequiredResult without SSE. Load balancing over plain HTTP without sticky sessions.

Loose coupling and stateless scaling have long been the norm for brokers; MCP is only arriving at this now. Apache Kafka →Load balancing of stateless calls over HTTP is handled by the ESB/API gateway; MCP moves it into the protocol. MuleSoft →
21.05.2026
MCP Apps and Tasks moved into the Extensions framework with reverse-DNS IDs
The RC introduces Extensions (reverse-DNS IDs, capabilities, versioning): MCP Apps serves HTML in a sandbox iframe, Tasks offloads long-running operations. A narrow core, growth through modules.

Long-running jobs, queues and versioned node plugins are the n8n model; MCP cements this in the spec. n8n →
21.05.2026
W3C Trace Context in _meta and a formal deprecation policy
The RC carries W3C Trace Context in _meta for end-to-end trace correlation and introduces a lifecycle policy: at least 12 months between deprecation and removal.

Collecting distributed traces via OTel/APM is a built-in Elastic feature; MCP merely passes the context through. Elasticsearch →
25.11.2025
Specification 2025-11-25: CIMD instead of DCR, Cross App Access
Instead of the problematic DCR (SEP-991) — Client ID Metadata Documents (client_id as a URL pointing to JSON). Cross App Access (SEP-990) shifts authorization to the organization's IdP.

Authorization and SSO are the reason teams deploy an LLM/Security Gateway; MCP covers part of it, while policies stay in the gateway. LLM & Security Gateway →