xenonstack-logo

Interested in Solving your Challenges with XenonStack Team

Get Started

Get Started with your requirements and primary focus, that will help us to make your solution

Please Select your Industry
Banking
Fintech
Payment Providers
Wealth Management
Discrete Manufacturing
Semiconductor
Machinery Manufacturing / Automation
Appliances / Electrical / Electronics
Elevator Manufacturing
Defense & Space Manufacturing
Computers & Electronics / Industrial Machinery
Motor Vehicle Manufacturing
Food and Beverages
Distillery & Wines
Beverages
Shipping
Logistics
Mobility (EV / Public Transport)
Energy & Utilities
Hospitality
Digital Gaming Platforms
SportsTech with AI
Public Safety - Explosives
Public Safety - Firefighting
Public Safety - Surveillance
Public Safety - Others
Media Platforms
City Operations
Airlines & Aviation
Defense Warfare & Drones
Robotics Engineering
Drones Manufacturing
AI Labs for Colleges
AI MSP / Quantum / AGI Institutes
Retail Apparel and Fashion

Proceed Next

Guides / Model Context Protocol

Model Context Protocol for the enterprise.

This guide covers what MCP is, how it works, where it falls short in a large organisation, and how to run MCP servers with the permissions, policies and audit trail your security team will ask for.

Read the guide MCP server development

From XenonStack, which is joining Anthropic's Claude Partner Network.

tools/list → erp-server policy: bounded 
{
  "name": "place_invoice_hold",
  "description": "Place a payment hold
     on a supplier invoice",
  "inputSchema": {
    "type": "object",
    "properties": {
      "invoice_id": { "type": "string" },
      "reason":     { "type": "string" }
    },
    "required": ["invoice_id", "reason"]
  }
}
Scopeerp.invoices:write (holds only)
BoundaryInvoices above the set limit go to a controller
TraceCaller, inputs, policy and result logged
What is the Model Context Protocol?

One standard way for AI to reach your systems. Not yet a way to govern them.

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in November 2024, that lets AI applications connect to external tools and data through a common interface. An MCP server exposes tools, resources and prompts; any MCP-compatible client can discover and use them, without a custom integration for each model.

For enterprises, MCP turns many separate model-to-system integrations into one reusable layer. Yet it also creates a new access path into systems of record. Who may call which tool, under which limits, and who answers for the result are questions the protocol leaves to you.

01 / How MCP works

Hosts, clients and servers

MCP uses JSON-RPC 2.0 messages. Servers run locally over stdio or remotely over streamable HTTP.

HOST

The AI application

The app the user or agent runs in, such as a desktop assistant, an IDE or an agent runtime. It manages one or more clients.

CLIENT

One connection

Each client keeps a one-to-one session with a server, negotiates capabilities and passes requests and results back to the host.

SERVER

Your system, exposed

A small service in front of a system, such as an ERP, CRM or warehouse, that publishes what the AI may use.

Tools

Actions the model can call, such as creating a ticket, querying an order or placing a hold. Each has a name, a description and an input schema.

Resources

Data the application can read into context: files, records, documents and query results, addressed by URI.

Prompts

Reusable templates a server offers for common tasks, so users and agents start from approved instructions.

02 / MCP vs APIs vs A2A

Where MCP fits next to what you already use

APIs and function calling Model Context Protocol Agent-to-agent (A2A)
ConnectsOne app to one serviceAny AI application to tools and dataOne agent to another agent
DiscoveryRead the docs, write the codeClients list tools and resources at runtimeAgents advertise skills to each other
Reuse across modelsRebuilt per model or frameworkBuild a server once, use it with any clientFramework-neutral by design
Best forFixed, well-known integrationsGiving agents governed access to systemsHanding work between agents
GovernanceYour API gatewayLeft to the implementerLeft to the implementer

They are complementary. MCP servers often wrap existing APIs, and agents that talk over A2A still use MCP to reach systems. The gap common to all three is governance, which is what the rest of this guide covers.

03 / Enterprise MCP

Five questions your security team will ask

An MCP server is a new door into a system of record. Before it reaches production, each of these needs an answer.

01
IDENTITY

Who is calling?

OAuth-based authorisation for remote servers, tied to your identity provider, so every call maps to a user or an agent.

02
SCOPE

What may it touch?

Permission scopes per tool, not per server, so "read orders" never quietly includes "refund orders".

03
INJECTION

Can content hijack it?

Emails, tickets and documents can carry instructions. Untrusted content must never be able to unlock a tool on its own.

04
AUDIT

Can we prove what happened?

A record of every call: caller, inputs, policy applied, approver and result, kept as long as your regulator needs.

05
REGISTRY

Which servers are approved?

A gateway and registry of vetted servers, so teams don't connect agents to unreviewed code from the internet.

CONTEXT

Resources carry business meaning

Data arrives with its owner, sensitivity and freshness, so the agent knows what it is reading and how far to trust it.

CONTROL

A decision boundary on every tool

Limits, approvals and blocked combinations are checked before a tool runs, not reviewed after.

ACTION

Every call leaves a decision trace

What the agent saw, which policy allowed it and who approved it, recorded in one place.

04 / Enterprise MCP servers

The servers enterprises build first

ERP

SAP or Oracle: read purchase orders, match invoices, place holds and draft journal entries for approval.

CRM

Salesforce or Dynamics: account context, opportunity updates and case history, scoped by territory.

Data warehouse

Snowflake or Databricks: governed, read-only queries with row-level security and source citations.

ITSM

ServiceNow or Jira: triage incidents, enrich tickets and run approved runbook steps.

05 / Governed MCP on ElixirClaw

Every MCP call goes through one gateway

ElixirClaw places a policy layer between your agents and your MCP servers. Teams keep the speed of MCP; security keeps one place to set the rules and read the record.

See the agentic AI platform

01

MCP gateway and registry

Only approved servers are reachable; each is versioned, reviewed and owned.

02

Policy per tool

Scopes, limits and approval rules attach to individual tools and are checked on every call.

03

Trust graduation

New tools start in shadow mode, then move to supervised, bounded and full autonomy as traces prove them.

ShadowSupervisedBoundedFull
04

Decision trace

One searchable record of every call, ready for audit, incident review and regulators.

06 / MCP server development services

We build the servers. You keep the control.

XenonStack engineers design, build and run MCP servers for your systems, delivered with the policies and traces above from day one. For agents on top, see our AI agent development services.

ASSESS

MCP readiness review

Map the systems agents need, rank tools by value and risk, and design the scopes and boundaries. Scope and timeline are agreed per engagement.

BUILD

Custom MCP servers

Production servers for ERP, CRM, data and ITSM systems, with OAuth, per-tool scopes, tests and documentation.

RUN

Managed MCP gateway

We operate the gateway, registry and policy layer on ElixirClaw, in your cloud, private cloud or on-prem.

FAQ

Model Context Protocol, answered

Who created the Model Context Protocol?

Anthropic introduced MCP as an open standard in November 2024. It is open source and supported by a growing range of AI applications and developer tools.

What is an MCP server?

A service that sits in front of a system and exposes tools, resources and prompts in the MCP format, so any compatible AI client can discover and use them.

Is MCP secure enough for enterprise use?

The protocol defines authorisation for remote servers, but scopes, approvals, injection defences and audit are up to the implementer. With a gateway and per-tool policy, it can meet enterprise standards.

How is MCP different from an API?

An API serves a specific application. MCP is a standard layer that AI clients use to discover and call tools at runtime, and MCP servers often wrap existing APIs.

How does MCP relate to agent-to-agent protocols?

They solve different problems. MCP connects an agent to tools and data; A2A connects agents to each other. Many enterprise designs use both.

Does MCP work only with Claude?

No. MCP is model-agnostic. Any client that implements the protocol can use an MCP server, which is why one server can serve many models and agents.

How long does it take to build an MCP server?

A read-only server for one system is usually quicker; servers that write to systems of record take longer because of scopes, approvals and testing. Timelines depend on the systems and risk profile—confirmed after readiness review.

Put MCP into production with control built in.

We'll map your first MCP servers, set the scope and boundary for each tool, and show you the gateway running in shadow mode.

Request a demo MCP server development