Give your AI a structural cortex and end-to-end encrypted memory through the Model Context Protocol — in Claude Desktop, Claude Code, Cursor, or any MCP client.
Claude Desktop is one click — no Python, no config files, nothing to install. Just add the extension and type your key. The Claude Code & Cursor steps further down are the developer route.
Four tools. Three of them only read; remember writes to your private,
passphrase-encrypted store that we cannot read.
alv_…. Treat it like a password.remember / recall). It encrypts your
notes on your device and never leaves it — we can't recover it. Use the same passphrase
on the website and in every client to share one memory. Skip it if you only use discover / verify.allivista.mcpb in (or double-click it).discover, verify, recall, remember appear in Claude.No Python or JSON needed — everything (server + dependencies) is bundled and runs locally on your machine.
Comfortable with a terminal and a config file? Read on. Otherwise use Claude Desktop above — same connector, zero setup.
pip install mcp httpx cryptography.mcp.json (project root) or ~/.claude.json{ "mcpServers": { "allivista": { "command": "python", "args": ["/path/to/allivista-mcp/api/mcp_server.py"], "env": { "ALLIVISTA_API_KEY": "alv_your_pro_key", "ALLIVISTA_MEMORY_PASSPHRASE": "your-memory-passphrase", "ALLIVISTA_BASE_URL": "https://api.allivista.com" } } } }
Restart Claude Code so it launches the server. On macOS/Linux use python3 and a full path like
/Users/you/allivista-mcp/api/mcp_server.py.
Same prerequisites as Claude Code (Python 3.10+, pip install mcp httpx cryptography, the unzipped server).
Add Allivista in Cursor → Settings → MCP → Add new server, or edit ~/.cursor/mcp.json:
{ "mcpServers": { "allivista": { "command": "python", "args": ["/path/to/allivista-mcp/api/mcp_server.py"], "env": { "ALLIVISTA_API_KEY": "alv_your_pro_key", "ALLIVISTA_MEMORY_PASSPHRASE": "your-memory-passphrase", "ALLIVISTA_BASE_URL": "https://api.allivista.com" } } } }
Reload Cursor; the green dot next to allivista means it connected.
Allivista is a standard stdio MCP server. Point any client at:
# command + args python /path/to/allivista-mcp/api/mcp_server.py # environment ALLIVISTA_API_KEY=alv_your_pro_key ALLIVISTA_MEMORY_PASSPHRASE=your-memory-passphrase # optional (memory) ALLIVISTA_BASE_URL=https://api.allivista.com
The server speaks MCP over stdin/stdout and proxies to api.allivista.com over TLS using your key.
| Tool | Access | Does |
|---|---|---|
discover | read-only | Hidden structural "bridge" concepts around a topic, from a 4.3M-paper cortex, with a constrained narration. |
verify | read-only | Checks whether the cortex's structure supports a claimed relationship between two concepts — a hallucination guard. |
recall | read-only | Finds a verbatim passage across your stored memory and which document it came from. |
remember | write | Stores a note verbatim in your private, end-to-end encrypted memory. |
You don't call these by hand — just ask your AI in plain language and it picks the right tool.
Discover hidden links — "Use Allivista to discover hidden bridges for Parkinson's disease."
→ concepts the literature connects only indirectly, plus a narration. Tip: pass comma-separated variants, e.g. alzheimer,alzheimers,dementia.
Fact-check a relationship — "With Allivista, verify whether amyloid-beta drives Alzheimer's."
→ a verdict (supported_direct … unsupported) + the bridging concepts — a hallucination check before you cite a link.
Remember something — "Remember this in Allivista: <your note>."
→ stored verbatim, encrypted on your device. Give it a title so you can overwrite it later.
Recall it later — "Recall from my Allivista memory: Project Halcyon vendor quote."
→ the exact passage + which note it came from. Search a distinctive literal phrase — recall is exact, not fuzzy.
See the full examples & cookbook →
Allivista runs on GeoLang — a language that encodes meaning as geometry: concepts as points in a structured space, related by their typed faces and co-occurrence bonds. The cortex (discover/verify) reads that structure directly — no embeddings. Your memory is a per-user keyed GeoLang dialect: words become opaque structural ids your machine can search but our servers can't read — verbatim and zero-knowledge at once. The cortex and memory are just two dialects; there's more. What is GeoLang? →
remember / recall data is encrypted with AES-256-GCM on your
device using a key derived from your passphrase. Allivista stores only ciphertext — we cannot read it, and cannot recover it
if you lose the passphrase.The Windows build isn't code-signed yet, so confirm your copy is genuine — the SHA-256 should match:
allivista.mcpb 47316b73a5385c2283e711b3f47ca0be4019a33431a7d86cc4673a9f2ead86be allivista-mcp.zip 4ace9622282c5bc2f673f5b6a5ddf10c1c36d316ba6c6a055881a21800e81c33
Check it — PowerShell: Get-FileHash .\allivista.mcpb -Algorithm SHA256 · macOS/Linux: shasum -a 256 allivista.mcpb