# What Is a Knowledge Base for AI Agents?

A [knowledge base for AI agents](https://www.knowman.io/guides/ai-knowledge-base) is a library the agent can search and read before answering. Tools let it find relevant material, fetch full [saved items](https://www.knowman.io/guides/how-to-save-articles), and ask questions across the library. This gives the agent your collected sources to work from, instead of relying only on information in its training data.

## What do agents need from a knowledge base?

An agent needs a way to find relevant material, inspect the source, and use that evidence to answer a question. Search identifies likely matches. Fetch opens the full text of a selected item. Ask lets the agent pose a question across the library and receive an answer with citations.

These are different jobs. A search result can point the agent toward an article, but the agent may need the full text to understand its argument or verify a detail. A cited answer can help with a broad question, while opening the cited item lets the agent check the evidence directly.

A [personal knowledge base](https://www.knowman.io/guides/personal-knowledge-base) gives the agent material you chose to save. In KnowMan, [saved articles](https://www.knowman.io/guides/ai-article-saver), newsletters, and feeds become part of that library. Articles tag themselves, so the collection can be searched without requiring you to organize each item by hand.

- Search: find saved items that match a topic or question.

- Fetch: read the full text of a saved item.

- Ask: get an answer from the library with citations to saved sources.

## How does an agent read a knowledge base through MCP?

MCP is a way for an [AI client](https://www.knowman.io/guides/connect-any-ai-to-your-notes) to connect to tools provided by another service. A connected client can call the knowledge base tools while working on a request. The agent can search, read, ask, or save a link according to the permissions set for its [agent key](https://www.knowman.io/guides/mcp-server-knowledge-base).

The agent uses those tools as needed; the library does not replace the agent's model or turn every answer into a web search. KnowMan answers come from saved reading, not the internet, and cite the saved sources so the evidence can be checked.

This connection works with Claude Desktop, [Claude Code](https://www.knowman.io/guides/claude-code-knowledge-base), Codex, Cursor, and any [MCP client](https://www.knowman.io/guides/what-is-an-mcp-server). The specific setup is done in the client by adding a configuration snippet. The agent key controls which actions that connected agent is allowed to take.

## What are the six KnowMan knowledge base tools?

KnowMan's MCP server has exactly six tools. Together, they let an agent work with saved reading and memories through distinct actions. The permissions chosen for an agent key determine which available actions the agent can use.

- kb_search: search everything you have saved.

- kb_get_item: read the full text of a saved item.

- kb_ask: ask a question and get an answer with citations.

- kb_memories: read your decisions, playbooks, facts, and notes.

- kb_recent: see what was saved recently.

- kb_save: save a new link to your knowledge base.

## How do you connect an agent to a knowledge base?

Connect an agent from an MCP client using an agent key and the configuration snippet provided for the client. Choose permissions that fit the intended use. Then ask a question that can be answered from something you have saved.

1. Open KnowMan Settings and create an agent key. Choose only the permissions you want that agent to have.

2. Copy the configuration snippet for your MCP client and add it to that client's configuration.

3. Use the connected client to ask the agent about a saved article, topic, or decision.

4. Check the citations in its answer and open a cited saved item when you want to inspect the source.

## What can an agent do, and what can it not do?

An agent with an agent key can search, read, ask questions, see recent saves, and save new links. Its access depends on the permissions selected when the key is created. This lets you give an agent the actions needed for a task without granting every available action.

Agents cannot write memories. They can read your memories through kb_memories, including decisions and the reasons behind them, but creating or changing those memories remains outside the agent's allowed actions. This distinction helps keep your recorded choices and notes under your control.

Before connecting an agent, consider what it needs to do. A research helper might need to search, read, and ask. A link-saving workflow also needs permission to save. Select the permissions that match the job, then test with a question about an item in your library.

## What else do people ask about agent knowledge bases?

A useful [agent knowledge](https://www.knowman.io/guides/ai-agent-memory) base contains sources the agent can retrieve when a question calls for them. The agent does not need to rely on an old chat to find your reading, because saved items and memories are available through their own tools.

- Can agents answer from the internet? KnowMan answers come from your saved reading, not the internet.

- Can an agent read the whole article? Yes. kb_get_item reads the full text of a saved item.

- Can an agent [remember decisions](https://www.knowman.io/guides/persistent-memory-for-ai-agents)? It can read memories, which keep decisions and reasons next to the reading behind them.

- Can an agent add a source? With permission, it can save a new link using kb_save.

## Connect an agent to the hosted KnowMan server

KnowMan also runs as a hosted remote MCP server at https://api.knowman.io/mcp over Streamable HTTP, so an agent can use the same six tools without a local process. In a client that supports MCP OAuth, add the server URL, sign in, and approve the KnowMan consent page. The page asks for the knowman.read scope, which covers search, read, and ask, and knowman.write for saving links. Access tokens last one hour and refresh automatically.

You can also create an API key in KnowMan settings and send it as an Authorization: Bearer header. With Claude Code, the documented command is claude mcp add --transport http knowman https://api.knowman.io/mcp --header "Authorization: Bearer <key>". Cursor uses an mcp.json entry with the same URL and header, and local Claude Desktop runs the open-source client (https://github.com/Essatt/knowman-mcp) with npx -y @knowman/mcp.

To start, create a KnowMan account at https://app.knowman.io, then ask the assistant to search or save, for example "search my KnowMan for agent memory" or "save this URL".

## Questions

### Does an agent need to be trained on my saved articles?

No. A connected agent can search and read saved material through tools when it answers. The knowledge base gives it access to your sources without making the agent rely only on training data.

### Can an agent see the full text of an article?

Yes. The kb_get_item tool reads the full text of a saved item. The agent can also search the library and ask questions that return citations.

### Can an agent edit my KnowMan memories?

No. Agents can read memories through kb_memories, but they cannot write memories. An agent key can be set up with only the permissions you choose.

### Which clients can connect to KnowMan?

KnowMan works with [Claude Desktop](https://www.knowman.io/guides/teach-claude-desktop-your-knowledge), Claude Code, Codex, Cursor, and any MCP client. Setup uses an agent key from Settings and a configuration snippet in the client.

## Related guides

- [Persistent Memory for AI Agents](https://www.knowman.io/guides/persistent-memory-for-ai-agents)

- [How to Build a Claude Knowledge Base](https://www.knowman.io/guides/claude-knowledge-base)

- [How Can AI Remember What You Read?](https://www.knowman.io/guides/ai-that-remembers-what-you-read)

- [How Do You Connect AI to Your Notes?](https://www.knowman.io/guides/connect-any-ai-to-your-notes)

- [What is an MCP server?](https://www.knowman.io/guides/what-is-an-mcp-server)

## More in this topic

- [Notion MCP server vs KnowMan](https://www.knowman.io/guides/notion-mcp-server-vs-knowman)
- [How to Give Your AI Context from What You Read](https://www.knowman.io/guides/give-your-ai-context)

## Give an agent access to your reading

Create an agent key in KnowMan Settings, choose its permissions, and connect it from an MCP client.

[Start free](https://app.knowman.io/signup)

Every new account starts with 14 days of Pro. No credit card.

Source: https://www.knowman.io/guides/knowledge-base-for-ai-agents