How to use Claude with your company data

The gap between Claude's ability and your context

Claude, built by Anthropic, is remarkably good at reasoning, summarising and drafting. Ask it to explain a regulatory framework or restructure a paragraph and it performs well. Ask it what your Q3 pipeline looks like or what decision was made in last Tuesday's product sync and it has no idea. This is not a limitation of the model itself. It is a limitation of what the model can see.

Large language models are trained on broad public data. They do not have access to your Google Drive folders, your Slack channels, your Confluence spaces or your internal wikis. The knowledge that makes your organisation run, the context that separates a useful answer from a generic one, sits behind logins and permissions that no foundation model can reach on its own.

So the question is not whether Claude is capable enough. It is how to get your company's information into the conversation in a way that is secure, scalable and not dependent on someone copying and pasting context every time.

What people try first

The most common starting point is manual context. You open Claude, paste in a document or a few paragraphs from a meeting transcript, and ask your question. This works for one-off tasks. It does not work when the answer depends on information spread across five tools, three departments and two years of Slack history.

Some teams move to Claude's own enterprise offering. Claude for Work and Claude Enterprise provide workspace-level features, conversation sharing and administrative controls. These plans are meaningful steps forward, but they come with trade-offs. They are expensive at scale, they support a limited set of integrations, and they lock your team into a single model provider. If next month a different model handles a particular task better, you cannot switch without changing your entire workflow.

A third approach is to build custom integrations using the Claude API. Engineering teams can write retrieval pipelines that pull relevant documents and inject them into prompts. This is powerful but requires significant development effort, ongoing maintenance and careful handling of how LLMs use your files. Most companies do not have the bandwidth or the appetite for that kind of project.

Connecting Claude to everything through a single layer

The approach that scales without custom engineering is to place a connective layer between your company's data and the AI models you want to use. This is what Fabric is designed to do.

Fabric connects to more than 50 tools your team already uses, including Google Drive, Dropbox, Notion, Gmail, Slack, GitHub, Confluence, Jira, Figma and many others. Once connected, Fabric indexes your content semantically, which means it understands what your documents are about rather than just matching keywords. When you ask a question through Fabric's AI assistant, the system retrieves the most relevant pieces of context from across all your connected sources and provides them alongside your prompt.

The result is that Claude, or whichever model you choose, can answer questions grounded in your organisation's knowledge. You are not pasting context manually. You are not building retrieval pipelines. You are not locked into one provider. One Fabric subscription gives your team access to the latest models from Anthropic, OpenAI, Google, DeepSeek, xAI and others, all working with the same underlying pool of company data.

How retrieval-augmented generation works in practice

The technical mechanism behind this is retrieval-augmented generation, commonly called RAG. When you ask Fabric a question, the system does not send your entire document library to the model. Instead, it identifies the specific passages, messages, files and transcripts most relevant to your query, then includes those as context when the model generates its response.

This matters for two reasons. First, it keeps responses grounded in real information rather than the model's general training data. Second, it means your data is handled with minimal exposure, only the relevant fragments are processed, and nothing is used for model training. Fabric's security posture includes AES-256 encryption and CASA Tier 2 certification, which addresses the concerns many teams have about using AI on confidential documents.

Using Claude (or ChatGPT) through Fabric's MCP integration

For teams that want to use Claude's own interface while still accessing company data, Fabric supports the Model Context Protocol. MCP is an open standard that allows AI applications to connect to external data sources in a structured way. With Fabric's MCP server, you can use Claude's desktop application or API and have it pull context from your Fabric-indexed content.

The same principle works in the other direction too. If your team prefers ChatGPT for certain tasks, you can access your Fabric content, your Fabric memories and your connected sources from within ChatGPT. You do not have to choose one model's interface and abandon the other. Fabric sits underneath both, providing the same retrieval layer regardless of which model you are talking to.

This means you do not have to choose between your preferred model's native experience and your company's knowledge. The two work together. The model does the reasoning, and Fabric provides the retrieval layer that supplies the relevant context from your cloud drive and connected applications.

Bringing Claude into Slack with Fabric Tag

One of the most practical use cases is making Claude available where your team already communicates. Fabric Tag lets anyone in your workspace @mention Fabric directly in Slack or Discord. When someone tags Fabric in a channel or thread, it retrieves relevant context from your connected tools and responds using Claude or another model of your choice.

This turns Slack from a place where knowledge gets buried into a place where it surfaces on demand. A new team member can ask about a process in a channel and get an answer drawn from documentation, previous conversations and project files, without anyone having to stop what they are doing to explain. The AI does the retrieval work, and the human gets the answer in the tool they are already using.

For teams that rely on Slack as a primary communication layer, this is a significant shift. Rather than searching across fragmented tools, you ask a question in the flow of work and the system finds the answer across all of them.

Why multi-model access matters

Tying your organisation to a single AI provider carries risk. Models improve at different rates across different tasks. Claude may be the strongest option for long-form reasoning today, but another model might handle code generation or data analysis better next quarter. Pricing changes, rate limits shift, new capabilities emerge.

Fabric's approach treats models as interchangeable components rather than infrastructure commitments. Your AI knowledge base stays the same regardless of which model is processing the query. If you want to compare how Claude and another model handle the same question against the same company data, you can do that without rebuilding anything. This is a meaningful form of risk reduction that becomes more important as AI capabilities evolve.

Setting up Claude with your company data

The process is less involved than you might expect. You connect your data sources to Fabric, which involves authorising each application through standard OAuth flows. Fabric indexes your content, building a semantic map of what exists across your tools. From there, you can ask questions through Fabric's assistant, through MCP in Claude's own interface, or through Fabric Tag in Slack and Discord.

There is no code to write, no infrastructure to provision, no retrieval pipeline to maintain. The AI agents and search features handle the complexity of finding and surfacing the right information. Your team just asks questions and gets answers informed by everything your organisation knows.

Giving agents scoped access to your tools

Fabric's agent access features add another dimension to how Claude works with your data. Rather than giving an AI model blanket access to everything, you can approve a specific agent, such as Instinct, Muse, or Claude, to use a specific set of tools for a limited time. For example, you might approve an agent to check your Gmail for relevant messages while it researches a question, without granting it permanent or broad access to your inbox.

This scoped, time-limited approach means the AI can do more than just search your indexed files. It can reach into live tools on your behalf, pull current information and take lightweight actions, all within boundaries you set. The permissions expire, so there is no lingering access to worry about. It is a practical middle ground between giving AI no tool access at all and giving it the keys to everything.

For teams concerned about what data the AI can access, Fabric provides controls over which sources are connected and which content is indexed. This is important for organisations handling sensitive information, as it allows you to maintain clear boundaries around what data AI tools can collect and process.

What this looks like day to day

Once the connection is established, the daily experience is easy to describe. A product manager asks Claude about the status of a feature, and the response draws from Jira tickets, Slack discussions and design documents in Figma. A support lead asks about a customer issue, and the answer pulls from help desk tickets, internal knowledge base articles and relevant email threads. A new hire asks how the deployment process works, and the response synthesises information from Confluence pages, GitHub READMEs and recorded onboarding sessions.

None of these queries require the person asking to know where the information lives. That is the point. The search layer handles discovery across all connected sources, and the model handles synthesis. The person asking the question just gets an answer.

This also creates a compounding effect over time. As your team adds more content and connects more tools, the knowledge base grows. Documentation that would otherwise go stale gets surfaced in response to questions, which means it continues to provide value rather than sitting unread. Fabric's self-writing documentation features take this further by helping keep records current as your organisation evolves.

Frequently asked questions

Can I use Claude specifically, or does Fabric choose the model?

You can select Claude as your preferred model within Fabric's AI assistant. Fabric gives you access to the latest models from multiple providers, including Anthropic's Claude, and you can choose which one to use for any given conversation or set it as your default.

Does my company data get used to train Claude or other models?

No. Fabric does not use your data for model training. Content retrieved in response to your queries is processed to generate answers, but it is not retained by the model providers or used to improve their models. Fabric's privacy and security architecture is built around this principle.

How does Fabric's MCP integration work with Claude?

MCP, the Model Context Protocol, allows Claude's desktop application to connect to external data sources. Fabric runs an MCP server that Claude can query, giving it access to your indexed company data. You set up the connection once, and Claude can then retrieve relevant context from your Fabric workspace whenever you ask a question.

Is this different from Claude for Work or Claude Enterprise?

Yes. Claude's own enterprise plans provide workspace features and some integrations, but they are limited to Claude as the only model and support fewer data source connections. Fabric acts as a layer underneath any model, connecting to 50+ tools and giving you the freedom to use Claude alongside other providers without duplicating your data infrastructure.

What types of company data can Fabric connect to?

Fabric connects to cloud storage (Google Drive, Dropbox, OneDrive, Box), communication tools (Slack, Gmail, Microsoft Teams, Discord), project management (Jira, Asana, Linear, Trello), documentation (Notion, Confluence, Google Docs), code repositories (GitHub, GitLab), design tools (Figma) and many others. The full list covers more than 50 integrations.

How secure is it to give Claude access to company data through Fabric?

Fabric uses AES-256 encryption for data at rest and in transit, holds CASA Tier 2 certification, and does not send your data to model providers for training. Only the specific passages relevant to your query are included as context, which follows the principle of data minimisation. You maintain control over which sources are connected and indexed.

Can I use Fabric Tag to ask Claude questions in Slack?

Yes. Fabric Tag lets anyone in your Slack workspace @mention Fabric to ask questions that are answered using your connected company data. You can configure which model handles these requests, including Claude.

How does this compare to building a custom RAG pipeline with the Claude API?

Building your own retrieval pipeline gives you full control but requires engineering resources for development, hosting, indexing, chunking, embedding and ongoing maintenance. Fabric provides a managed alternative that handles all of these components, connects to dozens of data sources out of the box, and lets you switch between models without rebuilding your pipeline.

Can different team members use different AI models through Fabric?

Yes. Fabric supports multiple models, and individual users can choose their preferred model for different tasks. One person might prefer Claude for writing and analysis while another uses a different model for code-related questions. The underlying knowledge base remains the same for everyone.

Does Fabric work if my team already uses Claude directly?

It does. Fabric complements direct Claude usage rather than replacing it. Through MCP, you can continue using Claude's own interface while gaining access to your company data. For teams that also use Slack heavily, Fabric Tag adds another entry point. The goal is to make your existing tools more capable, not to add another tool you have to learn from scratch.

Related pages

• How LLMs use your files

• What is retrieval-augmented generation

• How to use AI on confidential documents safely

• How to build an AI assistant for your team on your own data

• How to give an AI agent access to your knowledge base

• How to make AI search your Slack history

• What data do AI tools collect

• How to create a shared knowledge base