Start free
Guide MCP and AI agents

Connect your knowledge base through KnowMan's MCP server

Updated 5 min read

The short answer

To connect your KnowMan knowledge base over MCP, create an agent key in Settings, choose its permissions, and copy the supplied configuration into a supported client. Then test a specific saved source. KnowMan provides six tools for searching, reading, asking, reading Memories, viewing recent saves and saving links.

On this page9 sections
  1. Choose a concrete first connection
  2. Prepare the source you will use for testing
  3. Create an agent key and use the supplied configuration
  4. Use the right tool for each part of the task
  5. Test retrieval and answers separately
  6. Keep saving links separate from keeping decisions
  7. Check plan boundaries before expanding the workflow
  8. Connect the hosted KnowMan server
  9. Questions people ask

Choose a concrete first connection

Begin with a task that has an answer you can check. Find a saved article, read a known decision, or ask a question about a small collection of sources. This makes your first connection useful and gives you a clear way to assess the result.

KnowMan lists Claude, Codex, Cursor and your own agents as clients. Its setup guidance specifically names Claude Desktop and Claude Code. Choose a listed client you already intend to use, then follow the configuration supplied by KnowMan.

Define success before you connect: the client retrieves the intended source, the text matches your saved item, and you can identify what supports the answer. Keep connection setup and answer quality as separate checks.

Prepare the source you will use for testing

Your first test needs recognizable content. Save an article through the browser extension or web app, then inspect the saved item. Choose something with a distinctive title, phrase or argument so you can tell whether the client found the right material.

KnowMan reads saved articles in full and tags them automatically. You can search saved text, highlights and notes. If your test concerns a decision, keep its reasoning in a Memory and identify the reading that informed it.

Write a short expected result for yourself before asking the agent. For example: the answer should identify the chosen approach, explain the recorded reason, and point to the relevant article. Use it as a checklist when you connect your own client.

Create an agent key and use the supplied configuration

KnowMan's documented flow starts in Settings. Create an agent key with the permissions you choose, then paste the supplied configuration snippet into your MCP client. Use the current snippet as your reference for connection details.

Match access to your intended work. A source-reading task and a link-saving task have different needs. Choose only the permissions needed for the workflow you are testing, and review that choice before broadening the task.

  1. Open KnowMan Settings and create an agent key.
  2. Choose permissions appropriate to the intended task.
  3. Copy the configuration snippet supplied by KnowMan.
  4. Add that snippet using your chosen client's configuration process.
  5. Test retrieval of the saved source you prepared.

Use the right tool for each part of the task

The server has six tools. Understanding their individual jobs makes it easier to request evidence, inspect results and recognize the boundary between reading existing knowledge and saving something new.

  • Use kb_search to search what you have saved.
  • Use kb_get_item to read a saved item's full text.
  • Use kb_ask to ask a question and receive an answer with citations.
  • Use kb_memories to read decisions, playbooks, facts and notes.
  • Use kb_recent to see recently saved material.
  • Use kb_save to save a new link.
  1. For your first research task, request a relevant search.
  2. Ask the agent to read the matching saved item.
  3. Request an answer that distinguishes source evidence from interpretation.
  4. Inspect the cited material before using the result.

Test retrieval and answers separately

First ask for a source you know exists: Find the article I saved about our review process and read its full text. Inspect whether the returned material matches your saved item. This checks a narrower question than whether a recommendation is persuasive.

Next ask a question whose answer appears in that source. Request the supporting passage and inspect it. Then try a decision question that needs both a Memory and an article, such as: What approach did I choose, and what reading supported it?

Finally, ask a question your saved material does not settle. Use this to inspect how the response handles missing evidence. Your review should distinguish an unsupported conclusion, an inference and a statement grounded in a retrieved source.

Check plan boundaries before expanding the workflow

Free includes AI assistant connection, Memories and keyword search. Free Ask is capped at five questions a day. Pro adds semantic search; on Free, semantic search falls back to keyword search. Feeds, newsletters and ingestion rules are Pro features.

Every new account begins with 14 days of Pro, with no credit card required. Use the trial to assess the features your workflow needs, then repeat representative searches on the plan you intend to keep. Avoid assuming a successful trial demonstrates the same search behavior on Free.

If a connection test fails, narrow the check. Confirm the saved item in KnowMan, review the supplied configuration, and inspect the permissions you selected. Record the exact failure before changing anything. A clear, repeatable test makes the next troubleshooting step easier to choose.

Connect the hosted KnowMan server

Create a KnowMan account at https://app.knowman.io. In an MCP client that supports OAuth, add https://api.knowman.io/mcp, sign in on the KnowMan consent page, and approve access. Or create an API key in KnowMan settings and send it as an Authorization: Bearer <key> header.

For Claude Code, configure the hosted server with claude mcp add --transport http knowman https://api.knowman.io/mcp --header "Authorization: Bearer <key>". For Cursor, add a knowman entry to mcp.json with "url": "https://api.knowman.io/mcp" and "headers": {"Authorization": "Bearer <key>"}. To run the open-source client locally with Claude Desktop, use npx -y @knowman/mcp; setup details are at KnowMan MCP on GitHub.

Questions people ask

Do I need to build my own MCP server?

KnowMan provides an MCP server. Its documented setup uses an agent key and a configuration snippet from Settings.

Which clients does KnowMan list?

Claude, Codex, Cursor and your own agents. The setup guidance specifically names Claude Desktop and Claude Code.

What can a connected agent save?

kb_save saves a new link to your knowledge base. Agents cannot write Memories.

How should I check whether the connection works?

Retrieve a recognizable saved item, inspect its text, then ask a question with an answer you can verify against that source.

Connect an agent to knowledge you can check

Start your 14-day Pro trial, save a source, and create an agent key in Settings. Test one clear question before expanding your workflow.

Start free

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