Contexta MCP

Persistent memory & knowledge graph for your AI assistant.
Keyword + vector + graph search in a single lookup.

What is Contexta

Contexta is a memory MCP server that remembers like a human does — by meaning, by words, and by relationships.

Most memory tools give you one flavor of recall: embedding search, a raw log, or a graph you pay extra for. Contexta fuses all three into a single query. Your agent can ask "what did we decide about pricing with Acme last quarter?" and get back the right meeting, the right decision, and the people connected to it — in one round-trip.

Under the hood, Contexta indexes your notes, messages, and documents as typed entities (people, projects, decisions, meetings, tasks) and links them into a knowledge graph automatically. Every retrieval combines:

Keyword search — exact phrases, names, IDs
Vector search — semantic meaning across languages
Graph traversal — follow relationships between entities

Why Contexta

Typed entities out of the box — people, projects, decisions, meetings, tasks are first-class, not blobs of text
Knowledge graph included — no paid tier gate on the graph layer
Triple retrieval in one call — no need to stitch together three different memory APIs
30-second setup — install via Smithery, sign in with OAuth, done
Per-user OAuth — your memory is private; the server authenticates every request individually
Multilingual — English and Russian equally supported

Use cases

Persistent memory for Claude, ChatGPT, and Cursor across sessions
Long-term context for AI agents working on ongoing projects
Personal knowledge base that actually understands who and what you mean
Research and meeting recall without manually tagging anything

Install

Via Smithery (recommended)

One-click install from the Smithery listing — Smithery handles the OAuth flow and proxies connections through its gateway.

Manual — Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows):

config.json

{
  "mcpServers": {
    "contexta": {
      "url": "https://mcp.contexta.cc/mcp"
    }
  }
}

Claude will walk you through the OAuth sign-in on first use. See examples/claude-desktop.json.

Manual — Cursor

Add a new MCP server in Cursor → Settings → MCP:

Name: Contexta
Transport: HTTP
URL: https://mcp.contexta.cc/mcp

See examples/cursor.json.

Manual — ChatGPT

In ChatGPT → Settings → Connectors → Add custom connector → paste https://mcp.contexta.cc/mcp. See examples/chatgpt.md for full instructions.

Authentication

Contexta uses OAuth 2.0 with dynamic client registration (RFC 7591) and PKCE. On first connection, your MCP client opens a browser window, you sign in with your Contexta account, and the client receives a per-user access token. Tokens are refreshed automatically.

No API keys, no shared credentials — every request runs as a specific user with that user's private memory.

Tools exposed

Contexta MCP exposes the following tools to your AI client (non-exhaustive):

search — keyword + vector + graph search across your memory
read_page — fetch a note or document by ID
create_page — save a new memory
update_page — edit an existing memory
graph_search — traverse the knowledge graph
list_tasks / create_task / update_task — task management
list_workspaces — available workspaces
and more — the full tool list is discovered automatically by your MCP client.

License

MIT — see LICENSE.

This repository is a public showcase for the hosted Contexta MCP service. The service itself is operated by the Contexta team at mcp.contexta.cc.

Contexta README