@cogniahq/mcp

End-to-end encrypted memory for Claude Desktop, Cursor, Zed, Cline, and any MCP-compatible client. Bring your own OpenAI key. Open source under MIT.

npx -y @cogniahq/mcp init YOUR_COGNIA_API_KEY

That prints the config block you paste into Claude Desktop, Cursor, or Zed. Restart the client and your AI assistant can query your Cognia workspace as a native tool.

What this is

A Model Context Protocol server that exposes your Cognia memory to any MCP client. Your assistant can search, retrieve, and act against decisions, notes, documents, and integrations you've connected to Cognia (Slack, Notion, Linear, Jira, GitHub, Drive, and more).

It is the client-side surface. The backend it talks to is end-to-end encrypted at rest with per-org keypairs (libsodium Argon2id); the database holds only ciphertext. With BYOK enabled, your prompts and embeddings go to your OpenAI account on your contract, not ours. Threat model is public at https://cogniahq.tech/security.

Related MCP server: MCP Enhanced Data Retrieval System

Tools exposed

Mutating actions are drafted first by default. Execution is a separate tool call so the assistant cannot silently change your external systems.

Install

Claude Desktop

// ~/Library/Application Support/Claude/claude_desktop_config.json  (macOS)
// %APPDATA%\Claude\claude_desktop_config.json                       (Windows)
{
  "mcpServers": {
    "cognia": {
      "command": "npx",
      "args": ["-y", "@cogniahq/mcp"],
      "env": { "COGNIA_API_KEY": "ck_live_..." }
    }
  }
}

Cursor

// ~/.cursor/mcp.json  (or .cursor/mcp.json in your repo for project-scoped)
{
  "mcpServers": {
    "cognia": {
      "command": "npx",
      "args": ["-y", "@cogniahq/mcp"],
      "env": { "COGNIA_API_KEY": "ck_live_..." }
    }
  }
}

Zed

// ~/.config/zed/settings.json
{
  "context_servers": {
    "cognia": {
      "command": { "path": "npx", "args": ["-y", "@cogniahq/mcp"] },
      "env": { "COGNIA_API_KEY": "ck_live_..." }
    }
  }
}

Restart your client. Your assistant now has Cognia tools.

Don't want to remember the JSON? Run npx -y @cogniahq/mcp init and it prints all three blocks with your API key inlined.

Getting an API key

1. Sign up at https://cogniahq.tech (free up to 5 seats, no card)

2. Open Settings → API keys

3. Create a key. Copy the ck_live_... value (shown once)

Environment variables

Use as a TypeScript client

The same package also exports a typed REST client if you want to build against the Cognia API without the MCP transport.

import { CogniaClient } from '@cogniahq/mcp/client'

const cognia = new CogniaClient({ apiKey: process.env.COGNIA_API_KEY! })

const hits = await cognia.search.query({ query: 'postgres migration decision', limit: 5 })
for (const hit of hits) {
  console.log(hit.title, hit.snippet)
}

Full surface: cognia.memories.list / iterate / retrieve / update / delete, cognia.search.query, cognia.actions.plan / retrieve / execute. Built-in retry with exponential backoff and Retry-After support.

Build from source

git clone https://github.com/cogniahq/cognia-mcp.git
cd cognia-mcp
npm install
npm run build
node dist/cli.js --help

Why open source

If you ship an MCP server that touches company knowledge, your security team will want to read it before approving the install. So we open-sourced ours. The encrypted backend, search pipeline, and admin console stay closed. The surface that runs on your machine is right here.

The integrations SDK and a small cognia memory CLI will follow on the same MIT terms.

License

MIT. See LICENSE.

Links

• Cognia: https://cogniahq.tech

• API docs: https://cogniahq.tech/developers

• Security and threat model: https://cogniahq.tech/security

• Pricing: https://cogniahq.tech/pricing

• Issues: https://github.com/cogniahq/cognia-mcp/issues