Saltar al contenido
Leer en español

Connect Claude to Nexus

How to connect your agency’s Claude to Nexus, what the agent does once connected and how to disconnect it. The connector also works with ChatGPT and other MCP-compatible assistants.

Written for Nexus 0.68.3

Nexus is only available in Spanish. Screen and button names are quoted as they appear in the app, with their translation in parentheses.

Before you start

  • A Claude account that supports connectors. Anthropic defines which plans include them (its plans). With another assistant, its provider defines them.
  • A Nexus account with the permission «Conectar el agente y definir su horario» (Connect the agent and set its schedule). It is granted per role, in Configuración → Equipo → «Roles» (Settings → Team → Roles). It gives access to Agente → Configuración (Agent → Settings).
  • The agent’s connection is approved by the owner or an administrator. Once approved, the assistant works as the agency’s agent, with the agent’s permissions.

The agency has a single agent connection, and all its routines go through it. A new connection replaces the previous one. Each person can also connect their own assistant, if their role allows it (see «Your own assistant» below).

Connect the agent with «Añadir a Claude»

  1. In Nexus, go to Agente → Configuración (Agent → Settings). In «Conexión del agente» (Agent connection), «Conector de esta agencia» (This agency’s connector) shows the connector name, «Nexus - Agente de la agencia», and its address. If you belong to several agencies, the name also shows the agency in parentheses.
  2. In the «Claude» tab, click «Añadir a Claude» (Add to Claude). Claude opens with the connector ready: confirm it and connect it. In Claude Team or Enterprise, an Owner of the organization adds it, with the link copied by «Enlace para Owner» (Link for Owner).
  3. Claude opens the Nexus authorization page, which shows a code. In Agente → Configuración a «Solicitud de conexión» (Connection request) appears at the top, with the program asking to connect as the agency’s agent and a code. Click «Aprobar» (Approve) only if the code matches the one on the authorization page; if you do not recognize it, «Rechazar» (Reject).
  4. The notice says «Conexión aprobada» (Connection approved). The authorization page continues on its own and returns to Claude. The block says «Conectado como agente de» (Connected as agent of) the agency, with the program, who authorized it, when, and its «Último uso» (Last used).

Connect from the Claude directory

  1. In Claude, open the connectors directory, find Nexus and connect it.
  2. Claude opens the Nexus authorization page, which shows an eight-character code. It is valid for ten minutes and can be used once.
  3. In Nexus, go to Agente → Configuración (Agent → Settings), type the code in «¿El asistente muestra un código de ocho caracteres?» (Does the assistant show an eight-character code?) and click «Conectar» (Connect). Typing it ties the connection to this agency and approves it.
  4. The authorization page continues on its own and returns to Claude.

With ChatGPT or another MCP-compatible assistant

Nexus recommends Claude. The connector also works with ChatGPT and other MCP-compatible assistants that authenticate with OAuth. The connector address is the same; what changes is where you paste it. In Agente → Configuración, choose your assistant’s tab:

  • ChatGPT: «Copiar dirección» (Copy address) and «Abrir ChatGPT» (Open ChatGPT). In ChatGPT, open Plugins, click «+» and choose «Add custom MCP server». Paste the connector address and choose OAuth as the authentication.
  • Otro cliente MCP (Other MCP client): «Copiar dirección» (Copy address). In your assistant or orchestrator, add a remote MCP server with the connector address and OAuth authentication.

Once connected, the connection is approved as with Claude: through the «Solicitud de conexión» or by typing the eight-character code.

Letting Nexus notify the agent

In «Activación del agente» (Agent activation), Nexus stores the assistant’s trigger: a Claude routine, a ChatGPT agent or, for another assistant, an https address that receives a signed webhook. With it, Nexus activates the agent when someone delegates a task to it, leaves it a note or answers one of its questions, within the «Horario de atención» (Service hours) set by the agency. For the ChatGPT agent, the screen states that ChatGPT Business or Enterprise is required. The steps are in the manual, in Spanish: Conectar el asistente de la agencia.

Your own assistant

Whoever has the permission «Conectar su propio asistente» (Connect their own assistant) can connect their Claude, their ChatGPT or another MCP client in Agente → Mi asistente (Agent → My assistant), with their own connector, «Nexus - Agente personal».

  1. In «1. Añade tu conector al asistente» (1. Add your connector to the assistant), choose your assistant’s tab and add the connector as above.
  2. When you connect it, the assistant opens a Nexus page with a code. Type it in «2. Escribe el código de la página de Nexus» (2. Type the code from the Nexus page) and click «Conectar» (Connect). Typing it approves your own connection.

Your assistant sees and does the same as you, and what it registers waits to be confirmed. It is not the agency’s agent: it does not run the agency’s routines or handle delegated tasks. Whoever has «Conectar el agente y definir su horario» does not see «Mi asistente»: their assistant is the agent they configure. Without the permission, «Pedir ayuda al agente» (Ask the agent for help) offers «Nueva tarea» (New task), to delegate the work to the agency’s agent. The details are in Conectar tu propio asistente, in Spanish.

What the agent does

The agent is one more member of the team, named «Agente» (Agent), with its own permissions. It works with what the agency has in Nexus and with the flows the agency gives it: reading linked mailboxes, registering documents, updating the portfolio, handling tasks, drafting emails and quoting surety bonds, among others.

  • Nexus verifies what it registers. What the agent registers is verified by Nexus and, depending on the supervision chosen by the owner or an administrator in Agente → Supervisión (Agent → Supervision) («Revisar todo» (Review everything), the default; «Revisar excepciones» (Review exceptions); «Revisar solo lo bloqueado» (Review only what is blocked)), a person confirms everything or only what Nexus could not verify. Verification covers clients and policies; everything else it proposes, such as insurers, lines of business or criteria, always waits in Agente → «Por confirmar» (To confirm). If it corrects a record that was already confirmed, the change stays flagged until a person marks it as reviewed.
  • It applies some settings on its own. In these areas of the agency’s settings it makes the change without asking, within what the person who connected it or asked for the work can do: cancelling a policy, the agency’s presentation, publishing a flow, references, deadlines and renewal, document organization, criteria and assignment rules, and surety bond rules. Each change is listed in Agente → Permisos → «Historial de solicitudes» (Agent → Permissions → Request history), where it can be undone for seven days.
  • It asks for everything else. Deleting, inviting, a person’s data and vault accesses always ask; roles ask, unless someone gave it a permission for a limited time. The owner or an administrator can mark any area in «Qué pregunta siempre» (What it always asks) so that it asks there. Its requests arrive in the «El agente pide permiso» (The agent asks for permission) window, where a person approves them, returns them for correction or discards them. The details are in the manual, in Spanish.
  • It does not do what belongs to a person: sending an email (it leaves it to send, and it is sent by whoever has the permission «Enviar los correos que redacta el agente» (Send the emails the agent drafts)), opening the access vault or confirming its own proposals.
  • It works with the permissions of whoever asks. When it works for someone, for example on a task that person delegated to it, it cannot see or change what that person could not.
  • Everything goes into the activity log, with the agent as author and, if it acted on someone’s behalf, at whose request.

How the agent works, screen by screen, is in the manual, in Spanish: Cómo trabaja el agente and Verificación y supervisión.

Disconnect

  • The agency’s agent: in Agente → Configuración, in the connection block, «Cortar la conexión» (Cut the connection). The assistant stops working as the agency’s agent, and so do its routines. To connect it again, a new connection has to be approved.
  • Your own assistant: in Agente → Mi asistente, «Tus conexiones» (Your connections) shows each connected program; «Cortar la conexión» stops it from entering Nexus immediately. If you lose the permission, your connections stop working on their own.
  • Someone else’s: whoever manages the agent sees it in Agente → Configuración, in «Asistente de cada persona» (Each person’s assistant), and can cut it with «Cortar la conexión».

Removing the connector in the assistant does not cut the connection in Nexus: cut it there too.

Your data

What the agent reads through the connector reaches your assistant’s provider, Anthropic for Claude or OpenAI for ChatGPT, and is handled under your account and your terms with that provider. The details are in the privacy policy.

Help

Write to support: proyectomainapp@gmail.com.