One endpoint. Your vault, programmatically.

MCP — tool layer for AI clients

Endpoint: https://api.mindwiki.io/mcp. Claude (Pro), ChatGPT (Plus), Claude Desktop, Claude Code, and Codex are wired to connect over OAuth. Other MCP-aware clients work too — see Other MCP clients.

MCP exposes MindWiki as a set of tools (search, read, write, capture, ask, similarity, graph, lint, agent layer) that the AI client invokes through structured calls. The full list and argument shapes live in the API reference.

For MCP clients that can't handle remote MCP yet, the same endpoint accepts an API key as a query string — https://api.mindwiki.io/mcp?token=$MINDWIKI_API_KEY — as a token fallback. Same vault, same scopes.

REST — HTTP endpoints for scripts and automations

Base URL: https://api.mindwiki.io. REST exposes the vault as ordinary HTTP endpoints — /vault/pages, /vault/page/{path}, /vault/search, /vault/capture, /vault/ask, /vault/graph, /vault/health, /vault/export, and the /vault/changes sync log. Full reference in the API reference. Bearer auth:

curl -H "Authorization: Bearer $MINDWIKI_API_KEY" \
  "https://api.mindwiki.io/vault/search?q=onboarding"

Personal API keys are minted in Account → API Keys. Each key is named, has a creation timestamp, a last-used timestamp, and an explicit scope:

• read — search, read pages, list pages, ask, similarity, graph, vault health, stats, export.
• write — capture, create or update pages, delete pages, rebuild the vault index.

Keys start with mw_. The format is opaque — don't parse it. MindWiki stores only the hash; the full key is shown exactly once at creation. Lost it? Revoke and re-mint.

• AI clients with vault context. Connect Claude or ChatGPT to MindWiki via OAuth and your conversations start with the full set of your notes, decisions, and project specs available.
• Capture pipelines. Forward Readwise, email, Telegram, or webhook data into POST /vault/capture. Each one lands as a page under capture/.
• Notion / ClickUp / Linear mirrors. Send changed pages or tickets to POST /vault/page under their own folder prefix.
• Custom agents and scripts. Read-scope for retrieval-only tools; write-scope when the agent is authorized to capture or update pages.
• Backups + analytics. GET /vault/export gives you a clean zip of the entire vault at any time.

Search the vault

curl -H "Authorization: Bearer $MINDWIKI_API_KEY" \
  "https://api.mindwiki.io/vault/search?q=onboarding"

Capture from any external tool

curl -X POST "https://api.mindwiki.io/vault/capture" \
  -H "Authorization: Bearer $MINDWIKI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"title":"Readwise highlight","source":"readwise","content":"A saved highlight..."}'

Ask the vault a question

curl -X POST "https://api.mindwiki.io/vault/ask" \
  -H "Authorization: Bearer $MINDWIKI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question":"What recent notes mention agent-native memory?","max_pages":5}'

• Read: /vault/pages, /vault/page/{path}, /vault/search, /vault/graph, /vault/stats, /vault/health, /vault/export, POST /vault/ask.
• Write: POST /vault/capture, POST /vault/page, DELETE /vault/page/{path}, POST /vault/index/rebuild.
• Sync: /vault/changes, POST /vault/changes/ack (device clients).

POST /vault/page and DELETE /vault/page/{path} accept base_hash. If it doesn't match the current hash on the server, you get HTTP 409 with { "code": "CONFLICT" } and the actual current hash — pull, reconcile, retry. This is how the macOS sync engine avoids silent overwrites.

Per-minute limits scale by plan and bucket (sync, read, write, expensive). Responses include X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Bucket headers. On 429 you get a Retry-After back.