systems
Sistemas
erp · crm · sql
Kodex
Technology solutions
We give your AI a secure port to ERP, CRM, and SQL. Business sees decisions on live data; engineering sees MCP servers, tools, scopes, and audit.
What it is and why it matters
MCP (Model Context Protocol) standardizes how a model discovers tools and data. Stop wiring every chatbot to every API: one protocol, clear policies, auditable execution.
Systems, APIs, and actions the AI must be able to read or execute — with owner and risk.
MCP servers with tools and resources: the AI sees capabilities, not loose endpoints.
Scopes, approval gates, and logs: writes only with policy; reads with ACL.
Without MCP
With Kodex MCP
Live flow
Two lanes: tool discovery and live execution with policy — from agent to client system.
Flujo MCP vertical: descubrimiento de tools y ejecución con scopes hasta la acción en sistemas
01 · Descubrimiento
systems
Sistemas
erp · crm · sql
mcp server
MCP Server
tools · resources
catalog
Catálogo
schemas · versions
02 · Ejecución en vivo
agent
Agente / LLM
host · session
mcp client
MCP Client
call · stream
policy
Scopes / HITL
allow · approve
action
Acción
read · write · audit
Depth
Three depths. Start with safe read access and grow to governed execution on critical systems.
01
Nivel 01
Live data in chat — no writes
Ideal for querying stock, accounts, or tickets from the assistant without touching write systems yet.
Qué incluye
Ideal para · Teams that want real context in the assistant now.
Resultado · Answers anchored to live ERP/CRM.
02
Nivel 02
Read and act with scopes
Bounded write tools (create ticket, update field) with policies and human review where needed.
Qué incluye
Ideal para · Ops and CX teams that want less manual work.
Resultado · Fewer clicks · more actions completed by the agent.
03
Nivel 03
RecomendadoMulti-system + ISO audit
Full layer: multiple backends, tool discovery, observability, and deployment inside your perimeter.
Qué incluye
Ideal para · IT and architecture teams that need executable AI at scale.
Resultado · One port, many models — without rebuilding integrations.
MCP does not replace your APIs: it exposes them as discoverable, governed capabilities for AI.
MCP port
From agent to ERP: tools discovered, scopes applied, action logged. Business sees agility; engineering sees protocol and governance.
What the client gains
MCP is not ‘another API’: it is how AI uses your systems without explosive technical debt.
01
Quotes, stock, and tickets reflect real state — not yesterday’s export.
02
A reusable port for new models or channels, instead of N fragile integrations.
03
Scopes and HITL stop an agent from ‘improvising’ in ERP or billing.
04
Every tool call is recorded: defensible for IT, ISO, and leadership.
Dual reading
Same architecture, two languages.
Negocio
AI queries ERP without leaving chat
Ingeniería
MCP tool → API/ERP adapter → response schema
Negocio
It can only do what is allowed
Ingeniería
OAuth/scopes · allowlists · approval gates
Negocio
We switch models without rebuilding everything
Ingeniería
Stable MCP host · versioned servers
Negocio
We know what the agent did
Ingeniería
Structured logs · correlation IDs · replay
From fragile integrations to a stable port
Illustration: how effort redistributes when you stop wiring every channel to every API.
Casos de uso
01
Business: orders and logistics in chat. Technical: read/write tools over the ERP API.
02
Business: quotes based on the current customer. Technical: sync tools + post-action events.
03
Business: your internal software usable by AI. Technical: MCP server over internal APIs.
Information without real-time access is dead information. Static integrations are debt that grows with every AI channel.
Business
AI operates on real stock, payments, and customers — with less waiting for ‘IT to build another connector.’
Engineering
MCP servers, tool schemas, and scopes: discoverable, versionable, and auditable over your current APIs.
Illustration · Integration time
−60%
illustrative effort for new AI channels vs point-to-point connectors
↓ 60 % vs point-to-point APIs
↑ 45 pts vs point-to-point APIs
Port reuse index (0–100)
Illustrative example: the MCP port raises reuse when adding models or channels without rewriting backends.
Each lever shortens the path from ‘AI knows it’ to ‘the system does it’ — with control.
New channels reuse the same MCP port.
Queries against real ERP/CRM state, not copies.
Scopes and HITL on high-risk actions.
Every tool call is logged and correlated.
Cycle in days
12d → 3d
illustrative time to expose a new system to the assistant
Antes
Con Kodex
Antes
Con Kodex
Antes
Con Kodex
Antes
Con Kodex
Illustrative values. In project we measure your integration backlog and the post-MCP-port delta.
How we deliver
From system inventory to a production port — with clear deliverables for business and engineering.
Fase 1
What the AI must be able to do
We inventory systems, APIs, and critical permissions — with a read vs write risk map.
Fase 2
Latency, contracts, and scopes
We validate security, latency, and data contracts before exposing capabilities to the model.
Fase 3
Servers, tools, and governance
We deploy MCP servers, tool schemas, and access policies with observability.
Fase 4
When the stack changes, the port holds
We evolve connectors and policies as ERP/CRM or new models change.
Trust perimeter
Executable AI is an extension of your security policy — not an ownerless webhook.
Permissions
Each tool declares what it can do; AI does not inherit ‘admin’ by default.
Least privilege
Risk
Critical actions go through human approval or strict rules.
Approval gates
Evidence
Who, what, when, and on which system — correlated and retained.
Audit trail
No. MCP sits as a discovery and safe-execution layer over your existing APIs and systems.
With scopes, approval gates, and action audit — AI does not write without explicit policy.
Yes. We can deploy the MCP port inside your perimeter with the same ISO controls.
Yes. The port’s value is reusing servers/tools with Claude, GPT, or other MCP-compatible hosts.
Filtro de oro
Design, deployment, and governance of critical connectors with traceability.
Book MCP architectureStart with patterns and technical docs in KodexLAB.
View MCP guidesProcesos
Automatizaciones orquestadas
Menos horas y errores para negocio; eventos, MCP y logs para ingeniería — no triggers aislados.
Ver →RAG
Sistemas RAG
Base vectorial + citas: menos alucinaciones, más casos cerrados y conocimiento que mueve CRM/tickets.
Ver →LLM
LLM & asistentes de IA
Puestos aumentados: LLM + RAG + MCP con guardrails — respuestas con evidencia y acciones gobernadas.
Ver →Kodex MCP: your AI reads and acts on real systems — with scopes, audit, and without rewiring every channel.