Company Knowledge Base Agent

1. 01Knowledge is scattered across the organization.
2. 02Confluence is the central store — but finding the right page inside it is still slow.
3. 03Search across many spaces and pages, then give up and ask a colleague who already knows.

A ReAct agent — think, act, observe, repeat — connected to a Confluence MCP server. The server exposes Confluence as tools: search (text and CQL), get_page for full content, and list_spaces to know where to look. The agent decides which tool to call, reads what comes back, and keeps going until it has enough to answer, then answers with links to the pages it used.

Cut lookup time from about thirty minutes to a few, without building an index that has to be kept in sync.

Capabilities

• 01ReAct loop — reason, call a tool, observe, repeat
• 02search — full text and CQL
• 03get_page — full page content
• 04list_spaces — scope the search to the right space
• 05Answers returned with source page links

1. in

   Trigger

   A natural-language question from any employee.

2. 01

   Agent parses the question and selects the right Confluence MCP tool(s)

3. 02

   Calls get page, search page, search space or CQL query

4. 03

   Iterates with further calls when the first result is insufficient

5. 04

   Aggregates retrieved content and synthesizes an answer

6. out

   Output

   A direct, concise answer with links to the Confluence pages it came from.

Components

• Query interface
• Knowledge agent — Claude Sonnet 4.5
• Confluence MCP server
• Confluence spaces and pages (existing, untouched)

Model

Claude Sonnet 4.5

Tooling

Confluence MCP server

Retrieval

Live tool calls — no vector store, no separate index

Cloud

AWS (internal enterprise)

• 01

  Deciding when a single tool call is enough and when the agent needs to keep digging — over-querying is slow, under-querying is wrong.

• 02

  Serving every persona in the business, not just engineers, from one interface.

• 03

  Answering from documentation written by many teams, to no shared standard.

• Information lookup time fell from about thirty minutes to about five.

• Deployed enterprise-wide across all personas, not limited to technical teams.

Written as a de-identified case study. The employer's identity, space and page names, and documentation contents are not disclosed. What is described here is the architecture pattern and the measured outcome — nothing proprietary to the environment it runs in.