Make Zibri.ai Work Your Way: A Guide to Custom Prompts, AI Agents, and MCP Integrations

By Finn

Executive Summary

Custom prompts, AI agents, and MCP integrations let you shape how Zibri.ai thinks, what it knows, and where it pulls information from. Together, they turn a general-purpose AI chat into a focused assistant that speaks your language, draws from your specific knowledge base, and cites exactly where each answer came from. This guide walks through each feature, explains how they connect, and gives you a clear path to using them today.

Introduction: Why Customization Matters

Out of the box, Zibri.ai is already useful. You can capture notes, upload documents, and ask questions against your vault. But the default experience is generic by design — it has to work for everyone before it works well for anyone in particular.

That changes when you start customizing.

A researcher and a consultant have different needs. A researcher wants dense, cited summaries. A consultant might want clean bullet points ready to drop into a client deck. Neither of those is the default, and neither requires technical skill to set up. The tools are already there. Most users just haven't found them yet.

Customization is also where the AI stops feeling like a novelty and starts feeling like a real part of your workflow. When the AI responds in your preferred format, draws only from the project you're working on, and pulls in data from the tools you already use, the gap between "asking a question" and "getting a useful answer" gets much smaller.

One important note before we get into the features: AI-generated content in Zibri.ai is assistive. It is not professional advice, and you are responsible for verifying what it produces. That's not a disclaimer to skip — it's a practical reminder that the AI is a thinking partner, not a final authority.

Custom Prompts: Shaping How Your AI Responds

The fastest way to improve your Zibri.ai experience is a custom system prompt. You set it once in your chat settings, and it shapes every response in that session.

A system prompt tells the AI how to behave before you ask your first question. Want responses in bullet points? Say so. Prefer formal prose with citations at the end? Write that in. Need the AI to stay concise because you're scanning on a deadline? That's a one-line instruction.

You don't need to know how to code. You're just writing a short instruction in plain language — the same way you'd brief a colleague before a meeting.

Some practical examples of what a system prompt might say:

• "Respond in bullet points. Keep each point under 15 words."
• "Write in formal report style. Use complete paragraphs and cite your sources at the end."
• "Be concise. If the answer is short, keep it short."

The prompt doesn't change what the AI knows. It changes how the AI communicates what it knows. That distinction matters. The underlying retrieval — pulling from your actual notes and documents — still happens the same way. You're just controlling the presentation layer.

If you work across multiple projects with different audiences, consider setting a different prompt for each vault-scoped chat. More on vault scoping in a moment.

AI Agents: Turning Your Vault into a Specialized Knowledge Assistant

Custom prompts adjust the AI's voice. AI agents go further — they turn a specific vault into a dedicated assistant that answers exclusively from that knowledge base.

An AI agent can be created from any vault. Once built, it draws only from the notes and documents in that vault, which means its answers are focused, relevant, and grounded in your proprietary content. You can keep an agent private or share it with others.

This feature is available on the Pro tier.

Think about what that makes possible. A consultant who has built a vault of industry research, client notes, and frameworks can create an agent that answers questions as if it has read everything in that vault — because it has. A writer working on a long-form project can create an agent trained on their research notes and drafts, then query it as they write.

The agent doesn't invent. It retrieves. Answers are grounded in your actual content using Retrieval-Augmented Generation (RAG), and each response shows which notes or documents it drew from. That citation layer is important — it tells you exactly where the answer came from, so you can verify it and trace it back to the source.

A few things worth knowing:

• Agents are vault-specific. One vault, one agent. If you want an agent for a different project, create a new vault and build from there.
• Sharing is optional. You control whether an agent is private or accessible to others.
• Accuracy is your responsibility. The agent reflects what's in your vault. If your notes contain errors, the agent will surface them. Review outputs before acting on them.

Integrations via MCP: Connecting External Tools

Your vault is only as useful as the content in it. MCP integrations let you bring in content from external tools without manual copying and pasting.

MCP stands for Model Context Protocol. It's the mechanism Zibri.ai uses to connect to external data sources. The workflow is straightforward: you provide an endpoint URL for an MCP server, browse the resources available on that server, and import what you need directly into a vault.

Once imported, that content becomes part of your knowledge base. The AI can retrieve from it, cite it, and include it in agent responses — the same way it treats any other note or document.

No specific third-party integrations are named in Zibri.ai's current documentation, so the specific tools you can connect will depend on what MCP servers are available and what endpoint URLs you have access to. The protocol is the standard; the connections you build are up to you.

This feature is particularly useful if you work with content that lives outside Zibri.ai — internal databases, external research feeds, or structured data sources. Instead of manually copying that content into notes, MCP lets you pull it in directly and keep your vault current.

Scoping Chats to a Vault for Focused Answers

By default, Zibri.ai can draw from your entire knowledge base when answering a question. That's useful for broad queries. It's less useful when you're deep in a specific project and want answers that stay on topic.

Vault-scoped chats solve that. When you scope a chat to a specific vault, the RAG engine retrieves only from that vault. The result is tighter, more relevant answers — and fewer situations where the AI pulls in content from a different project that happens to use similar language.

Combined with a custom system prompt, a scoped chat becomes a focused work session. You've told the AI how to respond (the prompt) and what to draw from (the vault). That's a meaningful amount of control for a few minutes of setup.

The citation layer still applies. Every response shows which notes or documents it drew from, so you can trace the answer back to its source and verify it.

Putting It All Together: Example Use Cases

Here's how these features combine in practice.

Scenario 1: The Research Analyst

A research analyst is preparing a market briefing. They have a vault called "Market Trends" containing industry reports, clipped articles, and their own analysis notes. They've also connected an MCP server that feeds in structured data from an external source.

Their setup: 1. Custom system prompt: "Write in formal report style. Use complete paragraphs. Cite sources at the end of each section." 2. Chat scoped to the "Market Trends" vault. 3. MCP-imported content already in the vault.

They ask: "What are the three most significant shifts in consumer behavior from the last quarter?"

Zibri.ai retrieves from the vault, generates a structured response, and cites the specific notes and documents it drew from. The analyst reviews the citations, verifies the claims against the originals, and has a first draft of their briefing section in minutes.

On the Pro tier (20GB storage, 3,000 AI queries/month), they have enough headroom to run this kind of workflow regularly without hitting limits.

Scenario 2: The Consultant Building a Client-Facing Agent

A consultant has a vault of proprietary frameworks, case notes, and methodology documents. They want a way to query that knowledge quickly during client calls — without scrolling through notes.

They create an AI agent from that vault (Pro tier). The agent answers questions based exclusively on their methodology content. They keep it private. During a call, they can query the agent and get a sourced answer in seconds, then verify it before sharing it with the client.

Scenario 3: The Writer on a Long Project

A writer is working on a book. Their research vault contains hundreds of notes, source documents, and draft excerpts. They scope their chat to that vault and set a prompt that returns short, direct answers — no padding, no summaries they didn't ask for.

When they need to check a fact or find a connection between two ideas, they ask. The AI retrieves from their notes and shows exactly which ones it used. The writer verifies and moves on.

Personal tier users (5GB storage, 500 AI queries/month) can run this workflow comfortably for a single project. Researchers or analysts running multiple projects simultaneously may find the Pro tier's higher limits more practical.

Getting Started

You don't need to set up everything at once. Start with one feature, see how it fits your workflow, and add from there.

Step 1: Set a custom system prompt. Open your chat settings and write a short instruction describing how you want the AI to respond. Even a single sentence — "Be concise and use bullet points" — will noticeably change the output. This takes about two minutes.

Step 2: Scope your next chat to a specific vault. Pick the vault most relevant to your current project. Start a chat and limit it to that vault. Notice how the answers stay on topic.

Step 3: Add an MCP endpoint (if applicable). If you have access to an external data source that supports MCP, add the endpoint URL in your integration settings. Browse the available resources and import what's relevant into your vault.

Step 4: Build an AI agent (Pro tier only). If you're on Pro, pick a vault with enough content to be useful and create an agent from it. Decide whether to keep it private or share it. Test it with a few questions and check the citations it returns.

A reminder on accuracy: AI-generated content is an assistive tool. Zibri.ai's responses are grounded in your actual notes via RAG, and each response cites its sources — but you are still responsible for verifying what the AI produces before acting on it. The citations make that verification faster. They don't make it unnecessary.

These features are available now. Custom AI workflows are noted as coming soon, so expect the platform's automation capabilities to expand. For now, custom prompts, vault-scoped chats, MCP integrations, and AI agents give you substantial control over how the AI works for you.

Start with the prompt. The rest follows naturally.