If you use Obsidian with Claude Code and MCP, you have already assembled the architecture. The wall is the vault itself: it only knows what you wrote into it. Twenty years of email, iMessage threads, WhatsApp conversations, and every AI session you have ever had are invisible to it. So is the understanding that only comes from connecting them: who the people in your life actually are, what they care about, what your relationship with each of them looks like, what you should share with each, and which claims across your corpus are reliable enough to believe.
Obsidian is a thinking tool. It knows what you write, when you write it. That is the right design for notes. It is the wrong design for a personal AI that needs to reason about your relationships, your history, and the full arc of your working life.
Mandaire indexes the history you have already accumulated, without requiring you to write anything new.
| Source | Obsidian vault | Mandaire |
|---|---|---|
| Notes you deliberately write | Yes | Yes, via capture or extension |
| 20+ years of email (Gmail / iCloud) | No | Yes |
| iMessage history | No | Yes, via local bridge |
| WhatsApp threads | No | Yes |
| Apple Notes | No | Yes |
| Calendar history and attendees | No | Yes |
| ChatGPT conversation history | No | Yes, via export ingest |
| Claude and Gemini conversation history | No | Yes, via export ingest |
| LinkedIn connections | No | Yes |
| Photos (people, events, context) | No | Yes |
The structural difference: Obsidian is a tool for notes you actively author. Mandaire is a context layer built from the data trail your life already produces. They are compatible, not competing. If you have years of Obsidian notes, Mandaire can surface them alongside your email, messages, and AI sessions in a single MCP tool call. You do not have to choose.
Your vault may have notes about Maya Patel, your former collaborator. Your email has 340 threads. WhatsApp has a group where she appears under a different number. Your calendar has 12 meetings. Your Claude conversations reference "the Maya project" without naming her directly.
The Obsidian + Claude Code pattern cannot connect these. Each source is a separate island. When you ask "what do I know about Maya," you are asking against your notes only.
Mandaire resolves these into a single entity across sources, using a five-phase pipeline: canonical normalization, inverted-index matching, multi-feature scoring across email addresses, phone numbers, and display names, alias-overlap merge, and UUID assignment. A single entity survives alias drift across every source. The "Maya" in your iPhone contacts and the "[email protected]" in your email archive resolve to one person with a unified 20-year history. From that resolution, Mandaire builds the graph that answers the questions that actually matter: who Maya is in your life, what she cares about based on 20 years of interaction patterns, what your relationship with her looks like, what you typically share with her, and which claims about her across your sources are reliable enough to act on.
Even a well-assembled Obsidian vault with MCP access gives every AI tool the same view of your context. There is no concept of "Sam Nakamura is allowed to see my availability but not my financial situation" or "this reply to Jordan Chen should not surface information from a thread Jordan is not part of."
Mandaire has a disclosure layer built in. Every MCP response is filtered per the recipient, topic, context, and surface it is being requested for. You do not configure this by writing policy files; you configure it by using the system and letting it learn from your corrections.
Your AI gets the full context you have assembled. Every tool call returns the same view. You manage what you share in the prompt. This is not a failure of Obsidian; it is not designed for this problem.
Each MCP call is evaluated against who is asking, what they are asking about, in what context, and on what surface. The same question from two different people can produce structurally different responses, without you writing a policy for every case.
Obsidian is genuinely local-first: your vault is files on your machine. That is the right starting point. The question is what connects to those files and under what terms.
When you use Claude Code or any other AI client against your vault via MCP, that client sends your content to the AI provider. The context is local until it is not. This is not a criticism of Obsidian; it is the architecture of every AI client.
Mandaire runs on a private server under your control, with your data encrypted at rest. The AI client you use reads from Mandaire's MCP interface; it does not receive your raw email archive or message history. It receives a filtered, resolution-aware response. The raw corpus stays on your server.
This distinction matters if your context contains anything you would not want indexed in an AI provider's training pipeline or accessible via a compromised account.
Mandaire installs as an MCP tool in Claude, ChatGPT, or any compatible client. Your Obsidian workflow does not change. You add one MCP connection, and your AI can now see the 20 years of context your vault cannot reach.