Your business talks to AI. And AI talks to your systems.

Cord's MCP is no longer a one-way street. Cord is a server: an AI like Claude queries your AR and builds quotes using 7 tools. And Cord is a client: it connects to your CRM or ERP's MCP servers, under permissions you sign off on. You decide who touches what.

Start for free View plans

Available on all plans · uses the same API key (more active keys as you upgrade; live consumption is metered by use)

tools/call · cord.flouvia.com/api/mcp JSON-RPC 2.0

Claude · IA

cartera_vencida

Your CRM

→ tools/call "cartera_vencida"
→ get_saldo("El Zarco") · your CRM
"vencido": 412300, "dso": 47
✓ the AI drafts the quote with full context

bidirectional · with your permission7 tools

CORD AS A SERVER

7 tools. Two ways to connect.

Connect Cord to an assistant like Claude and the AI works with your real data: checks the pipeline, finds overdue accounts, looks up a client, or drafts a quote. There are two entry doors: JSON-RPC 2.0 over stateless HTTP at /api/mcp, where the 7 tools live; and an HTTP/SSE channel with session at /api/mcp/sse + /api/mcp/message. Both authenticate with your API key and respect each tool's scope.

The 7 tools run inside your org — the AI queries, it doesn't make things up
Write actions require a key with write permission
The SSE session is tied to your org_id: every query respects your RLS

Cord como servidor MCP JSON-RPC 2.0

HTTPSSE · streaming

listar_cotizacionesread

cartera_vencidaread

resumen_negocioread

buscar_clienteread

listar_productosread

detalle_cotizacionread

crear_cotizacion_borradorwrite

Mismo header, misma llave que la API7 herramientas

CORD AS A CLIENT

Cord also queries your client's systems.

Flip the arrow. You register the URL of your CRM or ERP's MCP servers and Cord connects to them as a client. When you build a quote with AI, the agent loop at /api/cotizaciones/ai-draft first asks those remote systems —a client's balance, last order, agreed conditions— and then builds the lines with that context. Up to 5 loops max, and it closes every connection when done.

Connects via SSE injecting each remote server's authorization token
Each external tool is prefixed by server to prevent collisions
The agent loop interleaves your tools and remote ones in a single conversation

Cord como cliente MCP Agent loop · máx 5

CRM del cliente

→ connect(https://crm.cliente.com/mcp)

→ listTools() → 4 herramientas

← get_saldo("El Zarco") → $412,300

✓ contexto listo · la IA cotiza

Consulta antes de cotizarCon tu permiso

AGENT GOVERNANCE

Each agent touches only what you sign off on.

No open access. Each org has a default agent, the Cord Assistant, and a permissions table dictating which external servers it can connect to. If a server isn't on its allowlist, the agent doesn't even see it. Everything lives with Row Level Security in the database: every query filters by your org_id and no one crosses data between businesses.

Register and toggle MCP servers in Settings › Developers
Grant or revoke the agent's permission per server, instantly
RLS on mcp_servers, agentes_ia, and agentes_permisos: org isolation

Gobernanza de agentes RLS · por org

Servidor A · ventas2 herramientas firmadas

permitido

CRM externoesperando tu firma

sin acceso

Servidor C · finanzasbloqueado por política

sin acceso

Cada acción queda en el audit logTú firmas

GOVERNANCE

Explicit access, never open

Every query respects your RLS

The MCP session is bound to your org_id: every tool the AI invokes —yours or an external server's— filters through Row Level Security. No agent ever sees another organization's data.

Every agent action lands in the audit log
Same API key for the REST API and MCP

Per-server permission, revocable instantly

Turn your agent's access to each external MCP server on or off from Settings, without waiting on a deploy.

Up to 5 agent-loop turns per quote

It's a safety cap: the AI queries your systems up to 5 times before drafting — never an infinite loop.

2 transports: stateless HTTP and HTTP/SSE

Use plain JSON-RPC for one-off calls, or the SSE channel with a session when you need continuous streaming.

Available on every plan

Inbound and outbound MCP share the same API key — there's no separate plan for connecting AI to your business.

FREQUENTLY ASKED QUESTIONS

What developers ask about Bidirectional MCP + agent governance.

Is Cord's MCP only for an AI to read my data?

No. Cord works both ways: as an MCP server (an AI like Claude queries your pipeline and builds quotes with 7 tools) and as an MCP client (it connects to your CRM or ERP's MCP servers under permissions you authorize).

Can any AI agent touch any data in my account?

No. Agent governance in Settings lets you define, server by server, which tools each agent can use — access is explicit, never open by default.

Do I need a different key for MCP and the REST API?

No, it's the same API key for both — the authorization header is identical.

REST API REST API Cord stops being just a screen for humans and becomes a system your other systems can talk to.