Mythos

📝MythOS MCP (📝Model Context Protocol) connects your MythOS knowledge library to AI assistants like 📝Claude, 📝Cursor, Windsurf, and any MCP-compatible client. It lets you search memos, read content, create and update memos, chat with your library via 📝Retrieval-Augmented Generation (RAG), explore communities, and sync changes — all from within your AI tools. This is the Memory layer of 📝The Augmentation Stack in production: structured, persistent, identity-aware knowledge accessible from any AI client on any device.

How It Works

Every MCP interaction begins with get_context — a tool that loads your Augmentation System memos before any memo is created or updated. This means your AI collaborator knows your formatting rules (Mythos Style), collaboration values (Mythos Soul), who you are (Mythos Human), and what your collaboration has learned (Mythos Memory) before it writes anything. This is what makes MythOS MCP identity-aware — the AI doesn't just access your notes, it knows who you are. See: 📝Collaborative Augmentation.

The four augmentation memos are private, editable memos in your library tagged #mythos-mcp-context. You can customize them at Settings > Augmentation or by telling your AI to update them during conversation. When your AI learns something about how you work, it can be stored in your Mythos Memory — and every future session, across every client, starts with that knowledge. See: 📝Implicit Loading.

What It Does

MythOS MCP exposes 27 tools to any connected AI assistant — covering memo CRUD, search, daily memos, tasks, templates, communities, library chat, and sync. All memo creates and updates via MCP trigger the same side effects as UI saves — 📝Knowledge Graph edge extraction, 📝RAG embedding updates, and newsletter notifications for substantive public changes.

It also exposes 8 resources (memo content, creator profile, memo feed, community metadata, and the four augmentation memos) and 3 prompts (enrich memo, summarize library, find connections).

For the complete tool reference with parameters, return shapes, and annotations, see: 📝MythOS MCP Tools.

For natural language prompt patterns, see: 📝MythOS MCP Prompts.

Distribution Channels

Claude.ai / Claude Mobile (OAuth — one-click setup)

MythOS is available as a connector in Claude. Authorize via OAuth and your library becomes accessible in any Claude conversation — including iOS and Android mobile apps. See: 📝How to Connect MythOS to Claude.ai.

📝Claude Code (CLI)

For developers using Claude Code in the terminal — one CLI command with your API key. See: 📝How to Set Up MythOS with Claude Code. For an automated setup script, see the 📝MythOS Claude Code Skill.

Claude Desktop / 📝Cursor / Windsurf (npm package)

For any MCP client that supports stdio transport, configure the mythos-mcp npm package (npmjs.com/package/mythos-mcp). See: 📝How to Set Up MythOS with Cursor or Claude Desktop.

Generate your API key at mythos.one/settings/api-keys (requires Scholar or Oracle tier).

Updating an Existing Install

Changes to get_context response content take effect on the next tool call — no client-side action required. Changes to the set of available tools (adding or removing a tool) require the MCP client to re-fetch its tool list. Most clients cache the tool list at connect time, so users must disconnect and reconnect (or restart the client) to see newly-shipped or newly-removed tools.

Technical Architecture

  • Server: Built with @modelcontextprotocol/sdk v1.27.1, Streamable HTTP transport
  • Auth: OAuth 2.0 (authorization code + PKCE) for Claude connectors, API key auth (x-mythos-key header) for Claude Code and the npm package
  • Backend: All tool calls route through the MythOS v3 API with per-user scoping. Creates and updates fire graph-edge extraction, RAG embedding, and newsletter triggers — matching UI-save behavior
  • Safety annotations: Every tool has readOnlyHint, destructiveHint, and idempotentHint annotations per MCP spec
  • Augmentation: get_context loads the creator's four augmentation memos (Soul, Style, Human, Memory) at the start of each memo-writing interaction. Managed via the Augmentation System dashboard at settings/augmentation
  • Templates: Discovered on demand via read_template (by templateId or fuzzy name) or list_templates rather than pre-loaded into every session. Template content uses curly-bracket {placeholder} syntax for agent-replaceable instructions
  • OAuth endpoints: Discovery at /.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource/api/mcp

Source Code

Related

Contexts

Created with 💜 by One Inc | Copyright 2026