Model Context Protocol
BlinkVault MCP
Use your own Claude, ChatGPT, Gemini, Cursor or VS Code subscription with your vault — answers come back with page-level citations. Works with any other MCP-compatible client Access is scoped, revocable and audited.
Connect your AI app
Most apps only need the server URL — you'll sign in to BlinkVault and approve access (read-only by default; tick "Make changes" to allow edits). Apps without sign-in support use an API key instead. On the Free plan MCP is read-only (search, fetch and read tools); Plus and Pro unlock tools that make changes.
https://blink.vaultica.io/api/mcp
- In Claude, open Settings → Connectors → Add custom connector.
- Name it BlinkVault and paste the server URL above.
- Click Connect, sign in to BlinkVault and approve.
Enable it from the tools menu in a chat, then ask: “Using my vault, what did the Q3 report say about revenue? Cite pages.”
Available tools
Read tools need read access; tools that change your vault need write access (the "Make
changes" box when connecting, or a key with vault:write).
| Name | Scope | What it does |
|---|---|---|
| search | vault:read | Find the most relevant passages, each with a title, snippet, page and a link back to the source for citations. |
| fetch | vault:read | Read the full passage or document behind a search result. |
| hybrid_search | vault:read | Advanced search with file-type / zone filters and pagination. |
| fetch_document_context | vault:read | Read the text just before and after a passage. |
| traverse_entity_relations | vault:read | Explore how people, places and topics in your vault connect. |
| search_vault · get_file · list_tags | vault:read | Find files, read file details and list tags. |
| upload_file · edit_foundation | vault:write | Save notes and files, rename, edit text or change sharing. |
| create_foundation · create_foundation_post | vault:write | Create share links and posts. |
| start_ingestion_job · update_zone_settings | vault:write | Import from cloud storage and change zone settings. |
Use tools/list on a live client for the authoritative,
machine-readable schema (input shape, descriptions).
Security & limits
- Sign-in connections use OAuth 2.1 with PKCE. Access tokens last one hour, are bound to this server only, and refresh tokens rotate on every use. See and revoke signed-in apps in Connected apps; revocation takes effect on the app's next request.
- Tokens are opaque, 256-bit random strings. Only their SHA-256 hash is stored — the plaintext is never recoverable after creation.
- Every call is logged (tool, status, latency, deny reason). You can revoke a token instantly above; revocation propagates to the validation layer in seconds.
- Tenant isolation is enforced server-side: a token can only access the issuing user's vault, regardless of arguments passed.
- Anomaly detection runs every 5 minutes and flags burst usage, repeated denials, or scope-mismatch patterns. Excessive abuse may auto-lock a token.
- Rate limits apply per user / per tool. Errors include a
Retry-Afterheader where relevant. - Default token lifetime is 30 days (configurable 1-365). Rotate regularly; never commit tokens to a repo or share them across machines.
Troubleshooting
- 401 Unauthorized: token is missing, malformed, expired, or revoked. Issue a new one above.
- 403 Scope mismatch: the token lacks the scope required by the tool.
Re-issue with broader scopes (e.g. add
vault:write). - Tool not found: the client is on an older protocol or the tool was
deprecated. Re-run
tools/list. - Empty results: if your vault was just connected, give the AI pipeline a few minutes to index new files before querying.
- Still stuck? See Support.
https://blink.vaultica.io/api/mcp · Protocol 2025-06-18 (also 2025-03-26, 2024-11-05) · OAuth 2.1