Second Brain

A Typemill tutorial details how to connect Anthropic's Claude to a self-hosted Typemill wiki through the Typemill MCP plugin, letting users save chat results as Markdown drafts and retrieve them in later conversations. The setup requires an HTTPS-reachable Typemill instance, the MCP plugin, and a Claude account that can add custom connectors; Claude authenticates via HTTP Basic auth using a Base64-encoded username:password sent as a custom authorization header to the endpoint https://your-site.example/api/v1/mcp. The guide recommends a dedicated editor or author account without publish and delete rights, with write tools set to "Allow after ask" and read tools to "Allow always," and notes that article text passes through the AI provider when Claude reads it.

Your Second Brain for Claude with Typemill Claude and ChatGPT have become daily tools, and a lot of valuable knowledge now sits in cloud platforms, buried in endless chat histories. This article shows how to connect Claude to Typemill via MCP. You can then save the best results of a conversation in your own self-hosted knowledge base, your second brain, and use them again in later chats. What you get - Save knowledge from Claude chats in your self-hosted knowledge base. - Use your knowledge base in new Claude chats. - Full control: Claude writes drafts, and you decide what gets published. What you need - A Typemill instance reachable over HTTPS . - The MCP plugin for Typemill. - A Claude account that can add custom connectors. The problem: good answers get lost in chats Three weeks ago, an AI assistant explained something in a way that finally made sense. You used it, moved on, and forgot where it was. Now you need it again, scroll through old chats, give up, and ask the same question from scratch. The knowledge was created, but it was not kept. The idea: save on demand, retrieve anywhere Typemill is a self-hosted wiki that stores its content as plain Markdown files. With the MCP plugin, Claude can work with your wiki directly from a conversation: 1. Talk. Explore a topic or solve a problem in a normal chat. 2. Save. When the result is worth keeping, tell Claude to save it to your wiki. 3. Review. Claude writes a draft. Nothing goes live until you read it and publish it. 4. Retrieve. In a later chat, ask what you learned about a topic. Claude searches your wiki and answers from your own notes. The conversation stays where it is. The knowledge moves to a place you can browse, edit, and trust. A wiki holds the result , not the history : one topic, one article, kept up to date. Why self-hosted and Markdown - Private by default. Claude connects through a dedicated user account with only the permissions you give it, for example an editor role without access to system settings. - You stay in control. Writing is draft-first, and deleting is a separate permission that you can leave switched off. - No lock-in. Articles are ordinary Markdown files in folders. If you switch to another assistant that supports MCP, your knowledge stays where it is. - Your server. Your knowledge base is stored on your own infrastructure. Keep in mind that when Claude reads an article, the text passes through the AI provider for that conversation. Step 1: Set up Typemill 1. Create a dedicated user. Give it the role editor or author without publish and delete rights and activate API access for it. A separate user keeps Claude's permissions limited and easy to revoke. 2. Note the password of that user. You'll need it in Step 2. 3. Enable the MCP endpoint. In the plugin settings, switch on: - Enable authenticated MCP endpoint - Read tools - Write tools optional, only if Claude should create or edit pages Step 2: Create the authorization header Typemill uses HTTP Basic authentication. Encode username:password in Base64 on your Linux machine: printf '%s' 'username:password' | base64 -w0 The result is your header value: Basic <encoded-string Step 3: Add the connector in Claude 1. Go to Customize → Connectors → Add custom connector . 2. Enter a name like "Typemill MCP Server" and the MCP URL of your site, for example: https://your-site.example/api/v1/mcp 3. Set authentication to None the OAuth options aren't needed here . 4. Add a custom header : - Name: authorization - Value: Basic <encoded-string 5. Name: 6. Save the connector and open its tool list. Check that all Typemill endpoints appear. 7. Set permissions per tool. A sensible setup: - Read tools list, get, find, search : Allow always - Write tools create, update, publish, unpublish, delete : Allow after ask This way Claude can browse freely, but you approve every change. 8. Read tools list, get, find, search : Step 4: Test it Start a new chat, make sure the connector is enabled for it, and ask: Check if you have access to the Typemill MCP server. Claude should find the tools and return basic site info such as the title, author, and base URL. Then try the read tools: List all pages on the site. Finally, check the write tools with a prompt like this: Write a short tutorial on how to connect Typemill with Claude via MCP and create a new draft in Typemill via the Typemill MCP server. Claude will ask for your permission before it writes. Afterwards, check in your Typemill instance that the article is there as a draft. Step 5: Use it in your daily work When a conversation produces something worth keeping, ask Claude to save it: Save the essence of this conversation as a short tutorial in my Typemill wiki. Check the existing structure first and update an article if the topic already exists. Later, in any new chat, use your notes as a source: Search my Typemill wiki and tell me what I have saved about reverse proxies. The more precisely you describe how your wiki is organized and how articles should look, the better the results. Put these instructions into your Claude project instructions or a skill. Tips and troubleshooting - Tools missing or 401 errors: Check that the Base64 string has no trailing newline, that the user has API access activated, and that the plugin options from Step 1 are on. - Writes are refused: Check the user's role permissions in Typemill's ACL settings. - Drafts: Content changes made through the API are saved as drafts. Nothing goes live until publish page is called, which is a useful safety net. - Deleting is permanent. Keep delete page on "Allow after ask", or leave it disabled. - Security: Always use HTTPS, since Basic auth sends credentials with every request. Don't share the encoded header, because it contains your password.