Skip to content
Connect your AI

Claude Code

Give Claude Code the specifications, architectural decisions and debugging knowledge that outlive a single checkout or terminal session. It can read those pages before coding and leave a handoff in the same vault used by your other agents.

Let Claude Code prepare setup

Paste this into Claude Code with its configuration-editing tools enabled. It can prepare the connection; you complete browser sign-in and any host approval.

Paste into your agent

Set it up yourself

Run this in your terminal:

bash
claude mcp add --transport http --scope user mindwiki https://api.mindwiki.io/mcp

Open Claude Code, run /mcp, choose mindwiki and authenticate in your browser. User scope makes the connection available across your local projects. For a repository-specific connection, choose project scope instead and review what will be shared with that repository.

Complete the MindWiki connection

On MindWiki's authorization page, MindWiki Email + Access Key is the recommended sign-in method. Existing email-and-password accounts can choose Legacy Sign In; Sign in with Apple remains available. All methods open the same account and vault—no replacement account is required. New users choose Create a new vault, save their generated access key and continue. Review the requested permissions and choose read-only access when that is all you need. Access Keys explains the supported methods.

Return to your agent after approving. Your sign-in key belongs on MindWiki's pages, not in chat or an MCP configuration file. The client handles its own OAuth credentials.

Check that it works

Ask your agent:

text
Use MindWiki to read my account status, then search for a note about one of my projects. Read a matching page and cite its title and path. Tell me if the vault is empty rather than making up a result.

Inspect the tool result and compare it with a page in MindWiki Cloud. For a new vault, the agent can read the operating guide and initialize the starter pages when account status permits it.

To check saving after granting write or capture access, ask “Save a new MindWiki capture titled Connection check with today's date, then read it back and give me its path.” The same file is available in your MindWiki apps.

Put project memory to work

  • Before coding: “Read the auth specification and past decisions in MindWiki, then summarize constraints before proposing a change.”
  • After debugging: “Save the observed cause, failed approaches and working fix as a troubleshooting note.”
  • Before a handoff: “Save the current goal, changed files, verification and remaining work so another agent can continue.”

Keep repository instructions in the repository when they belong to the codebase. Use MindWiki for cross-project knowledge and durable notes; connecting does not rewrite your CLAUDE.md or AGENTS.md files.

Troubleshooting

Use claude mcp list and /mcp to inspect the connection. If tools are missing after adding the server, start a new session. If sign-in was canceled or access revoked, authenticate again through /mcp. Use OAuth for interactive work; dedicated scoped API keys for headless jobs are covered in API Access.

Permissions and disconnecting

Available tools depend on the scopes you approved, your MindWiki plan and vault permissions, and the tools your client exposes. Free includes one external AI connection with core vault tools. See MCP tool access.

To disconnect just this client, revoke its grant in MindWiki Connections. Other agents keep their own connections. Never send access keys or OAuth tokens to support.

More help

MCP tools · Agent permissions · All AI agents

Setup reference: Claude Code MCP documentation, checked October 20, 2026.