Connector documentation

• A NEXUS account. Sign-up is open and free at https://nexus.eblas.link/en/signup: the Free plan («Gratuito») includes up to 150 new memories and 2 agents a month (details in section 8 of the Terms).
• A client that supports remote MCP over HTTP: Claude (web, desktop and mobile), Claude Code, ChatGPT, Cursor or others.

Nothing needs to be installed on your computer to use it from Claude or ChatGPT.

• In Claude, open Customize → Connectors and add a custom connector (or find it in the directory).
• Use this URL: https://nexus.eblas.link/mcp
• Click Connect. NEXUS opens so you can sign in with a code sent to your email.
• Review the permissions Claude requests and authorize. You can revoke access at any time from your NEXUS account.

From Claude Code:

claude mcp add --transport http nexus-agi https://nexus.eblas.link/mcp

Then run /mcp inside Claude Code to authorize in the browser.

• In ChatGPT, open Settings → Apps → Advanced settings and turn on Developer mode.
• Create a new app, name it “Nexus AGI”, use this URL: https://nexus.eblas.link/mcp and choose OAuth authentication.
• NEXUS opens so you can sign in. Review the permissions and authorize.

ChatGPT stores the tool list from the moment you create the app. If your agent later does not see a new NEXUS tool, refresh the app from the same screen.

Claude and ChatGPT have no hooks: nothing recalls memory on your agent’s behalf. When recalling depends on the agent judging it necessary, it rarely does; with a concrete order (every turn, this tool, this argument) it always does. Paste these lines at the top of its instructions:

Your long-term memory is NEXUS (connector "Nexus AGI").
Every turn, before answering, call retrieve_memories with my message as the query and time_range="all_time".
If a result carries duda_de_fusion, decide it with decide_merge_proposal before answering; don't ask me about it.
When you learn something you would want to know next time (a decision, a preference, a mistake), save it with process_observation in that same turn.

• ChatGPT: Settings → Personalization → Custom instructions (1500 characters; they fit next to what you already have).
• Claude: Settings → Profile → Personal preferences, or a project’s instructions if you only want NEXUS there.

NEXUS uses OAuth 2.1 with dynamic client registration and PKCE (S256). There are no tokens to copy and paste. The permissions you grant are:

• memoria (memory): read and write episodes, facts and procedures.
• grafo (graph): read and write entities and relations.
• cognicion (cognition): rules, hypotheses and reasoning queries.
• trabajo (work): goals and tasks.

For scripts or agents without OAuth, the account owner can issue an agent credential (nxa_…) in the console and send it as an Authorization: Bearer header.

Talk to your agent as usual; it decides which NEXUS tool to use. Illustrative examples:

• “Remember that in this project payment records are never deleted: they are archived, because accounting audits them.” → stores a rule.
• “Before changing the database migration, what did we decide and what went wrong last time?” → recalls earlier decisions and mistakes.
• “Create a goal: launch the beta on October 15, and tell me which tasks are still open.” → creates the goal and lists the tasks.
• “The decision to use the old message queue no longer applies: we use a different one now.” → marks the old memory as no longer valid, without deleting it.
• “What does NEXUS know about our billing provider?” → gathers facts, relations and episodes about that entity.

Read-only tools (query, list, search) are marked as such; tools that create data are marked as writes, and tools that overwrite or delete data are marked destructive, so your client asks you to confirm before using them. By area:

• Memory: store and recall episodes, facts and procedures; mark a memory as no longer valid.
• Graph: entities (people, projects, systems) and their relations, with validity over time.
• Rules and hypotheses: rules the agent should follow and hypotheses with their evidence.
• Goals and tasks: create, update and close goals and tasks.
• Agent messaging: lets your agents leave messages for each other.

On agents that run on your computer (Claude Code, Cursor), NEXUS offers optional hooks that recall memory when a session starts and save what matters when it ends. They are installed only if you ask: your agent can fetch the installer with the get_setup_kit tool and you approve it. It does not apply to Claude on the web or desktop, or ChatGPT: there, the instructions in section 4 play that role.

• NEXUS knows what it is told. Without hooks, it only has what the agent stores on purpose.
• A memory can become outdated. NEXUS tracks validity and lets you replace memories, but it cannot tell on its own that something in the world changed: correct it when it does.
• Automatic conclusions (hypotheses, suggested relations) can be wrong and are presented as such, with their evidence.
• Beta: tools may change and there may be maintenance windows.

Each account is an isolated tenant: one account cannot read another’s data. NEXUS does not read your chat history or the memory of Claude or ChatGPT; it only receives what your agent sends through its tools. We do not sell your data or use it to train third-party models.

Full details in the privacy notice: https://nexus.eblas.link/en/privacy

Email us at [email protected]. To report a security issue, use the same address with “Security” in the subject.