AI that can pull context from Confluence and Slack

Summary

• Yes. A Databricks AI assistant can pull context from Confluence and Slack through the Model Context Protocol (MCP), the open standard for connecting agents to external tools, documents, and workflows.
• Databricks provides managed MCP connectors for common workplace tools, including Slack and Confluence, so an agent can reference a wiki page or a chat thread at query time instead of having it pasted into a prompt.
• Every connection is governed through Unity Catalog and Unity Gateway: managed OAuth handles sign-in, permissions are enforced on each call, and every tool invocation is logged for audit.
• For documents, Agent Bricks Knowledge Assistant, now generally available, ingests unstructured content and makes it retrievable by any agent, with user permissions and metadata respected.
• The evaluation questions that matter: does the assistant connect through a standard protocol, does it enforce per-user permissions, and are both the data and the tools governed in one place?

AI that can pull context from Confluence and Slack
Yes. A Databricks AI assistant can pull context from Confluence and Slack, and it does so through the Model Context Protocol (MCP), the open standard for connecting agents to external tools, documents, and workflows. Instead of copying a wiki page or a chat thread into a prompt, the assistant connects to the source, retrieves what's relevant at query time, and grounds its answer in current company knowledge. Databricks ships managed connectors for these tools and governs every connection through Unity Catalog, so an assistant reads only what the person asking is allowed to see.
That last point separates a useful enterprise assistant from a demo. Pulling in Confluence and Slack is the easy part. Doing it while preserving permissions, keeping credentials out of prompts, and logging every access is the hard part, and it's where governance earns its keep.

What does pulling context from tools like Confluence and Slack mean?

Most of a company's working knowledge lives outside its databases: specs and runbooks in a wiki, decisions and incident threads in chat, tickets in a tracker. Pulling context means an assistant can reach into those systems of record and use their current contents to answer a question, rather than relying on what a user remembers to paste in. The right design connects to the source and fetches on demand, so answers reflect the latest content and respect who is allowed to read it. The alternative, dumping documents into a prompt, breaks down on freshness, scale, and access control.

What to look for in an AI that connects to your tools

Capability	Why it matters
Standard connection protocol	An open protocol like MCP gives one integration pattern for many tools, not brittle one-off connectors.
Managed connectors	Ready-to-use connectors for common tools cut setup and avoid custom code for each source.
Centralized authentication	Managed OAuth keeps credentials out of prompts and code, so secrets aren't scattered across apps.
Per-user permission enforcement	The assistant must return only what the requesting user is allowed to see, on every call.
Tool and scope selection	Admins should choose which tools and actions are exposed, rather than exposing everything by default.
Unstructured document retrieval	Wikis and files need retrieval that carries metadata and user constraints, not just keyword search.
Audit and usage logging	Every tool call should be recorded for security review and cost tracking.
One place to govern data and tools	Governing tables and tool connections together avoids gaps between the two.

How Databricks approaches this

On Databricks, agents and Genie connect to external context through MCP, and the platform supports three kinds of MCP servers: managed servers that Databricks hosts, custom servers you host as a Databricks App, and external servers you register as Unity Catalog MCP services.

• Managed connectors for workplace tools. Databricks provides managed MCP connectors for common tools, including Slack and Confluence, along with others such as Jira, GitHub, Google Drive, and SharePoint. Managed OAuth handles sign-in centrally, so agents reach these systems without exposing secrets.
• Genie and Agent Bricks use them. Genie can use these connectors to bring tool context into an answer, and Agent Bricks natively supports MCP so agents get governed access to APIs, databases, and SaaS tools through one interface. Available integrations are discoverable in a shared catalog.
• Documents through Knowledge Assistant. Agent Bricks Knowledge Assistant, now generally available, ingests unstructured documents and makes them retrievable by any agent, incorporating system context, metadata, and user constraints.
• Governed by Unity Catalog and Unity Gateway. MCP services are Unity Catalog securable objects. Grants control who can invoke each service, tool selection limits what's exposed, permissions are enforced on every call, and admins can apply service policies to allow or deny individual tool calls. Every invocation is logged.

See the Databricks documentation on MCP tools for agents and available managed MCP servers, plus the blog on the governed enterprise agent platform.

Use cases

Support and operations. A support agent answers a customer question by pulling the current runbook from Confluence and the latest incident discussion from Slack, so its response reflects what the team knows right now.
Analytics with document context. An analyst asks Genie a question that blends governed tables with a product spec in Confluence, getting a numeric answer plus the written rationale behind a metric.
Onboarding and internal Q&A. A new hire asks where a process is documented, and the assistant retrieves the relevant Confluence page and points to the Slack channel where it's discussed, all within the access they already have.

FAQs

Does the AI copy my Slack and Confluence data, or read it at query time?

With MCP, content is exposed through a connector and referenced when relevant, so the assistant retrieves what it needs at query time rather than requiring you to paste it into a prompt.

Will it respect who can see what?

Yes. On Databricks, MCP services are governed by Unity Catalog, permissions are enforced on each call, and users see only the tools and data they've been granted.

What other tools have managed connectors?

Beyond Slack and Confluence, Databricks offers managed connectors for tools such as Jira, GitHub, Google Drive, and SharePoint, and you can register other external MCP servers in Unity Catalog.

Do I need to build anything to connect a tool?

Managed connectors need no custom code. For a system without one, you register an external MCP server or host a custom one as a Databricks App, then govern it the same way.