Start with the knowledge your task actually needs
Choose one task before connecting anything. Perhaps you want to revisit an architecture decision, compare arguments from saved technical articles, or apply a review playbook. A concrete question makes it easier to judge whether the knowledge base supplies useful evidence.
KnowMan can hold saved articles, highlights, notes and Memories. Memories keep decisions and their reasons alongside the reading behind them. Connected agents can read that material, giving you a way to bring your own thinking into a task.
For a first exercise, choose a decision with a clear source trail. Write down the question, identify reading you expect to matter, and define what a satisfactory answer must include. Keep this small enough that you can inspect every source the agent uses.
Prepare a useful starting collection
Save relevant articles through the KnowMan browser extension or web app. KnowMan reads saved articles in full and organizes them with automatic tags. You can also search highlights and notes, which helps when your own annotation contains the detail you need.
Add a decision or playbook when you want the agent to understand your preferred approach. Make its scope explicit: what you decided, why, which sources informed it, and what would make you reconsider. These are writing recommendations for useful context.
- Save the reading relevant to one current task.
- Check that the saved items contain the passages you intend to use.
- Record your decision and reasoning in KnowMan.
- Write a question that asks for both the decision and its supporting sources.
Connect using the setup supplied in Settings
KnowMan lists Claude Code among its MCP clients. Its documented connection flow begins in Settings: create an agent key, choose permissions, then copy the supplied configuration snippet into your client. Use that supplied configuration rather than assembling an endpoint or authentication format from examples elsewhere.
Choose permissions for the task you intend to perform. If your immediate goal is research, begin with the access needed to retrieve material. Add saving access when you have a concrete reason for the agent to save links.
Keep credentials out of prompts, shared notes and screenshots. For your first connection check, ask for a specific saved item you can recognize. Successful retrieval gives you something concrete to inspect before attempting a broader task.
Understand what the six tools do
KnowMan exposes six MCP tools. Their roles help you write clearer requests and distinguish finding evidence from producing an answer. You do not need every tool for every task; start with the operation that matches your question.
For a decision question, request relevant Memories and supporting articles. For a source question, request search and full-item reading. Agents cannot write Memories. Saving a link through kb_save is a separate action from recording an approved decision and its reasoning.
- kb_search searches the material you have saved.
- kb_get_item reads the full text of a saved item.
- kb_ask answers a question with citations.
- kb_memories reads decisions, playbooks, facts and notes.
- kb_recent shows what was saved recently.
- kb_save saves a new link to your knowledge base.
Ask for evidence before a recommendation
Try this request: Search my KnowMan knowledge base for the decision about our review process. Read the relevant Memory and supporting articles. Explain what I decided, why I decided it, and which source supports each reason. Separate retrieved facts from your recommendation.
This gives the task an inspectable shape. You can check whether the answer found your actual decision, whether the reasoning matches the saved record, and whether a proposed change follows from the evidence.
When reading a cited answer, open the citations and inspect the original material. Check dates, scope and qualifications. A source may support one part of a recommendation without supporting every conclusion around it. Ask a narrower follow-up when the connection is unclear.
Keep the workflow clear as your reading grows
Use your knowledge base to carry durable context between tasks. A useful playbook states the outcome you want, the constraints that matter, and examples of acceptable work. Keep recommendations distinguishable from decisions you have actually approved.
When an agent finds a relevant new article, you can ask it to save the link through kb_save if its permissions allow that action. Review the article before treating it as support for a decision.
After completing a task, record any decision worth keeping in KnowMan yourself. Include the reason and the conditions for revisiting it. Because agents cannot write Memories, make this an explicit review step rather than assuming an agent's answer became a saved decision.
Know which plan features your workflow uses
Free includes AI assistant connection, Memories, keyword search and Ask capped at five questions a day. Pro adds semantic search, which searches by meaning. On Free, semantic search falls back to keyword search. Feeds, newsletters and ingestion rules are Pro features.
Every new account starts with a 14-day Pro trial, with no credit card required. After the trial, you can remain on Free. Evaluate the workflow against the plan you expect to keep, especially if finding articles by meaning matters to your tasks.
Start with one repeatable question and a known set of sources. Expand when you can reliably inspect what was retrieved, explain why it matters, and recognize when the knowledge base does not contain enough evidence.