Raycast MCP plugin for Neocortex-powered long‑term memory.
This package provides a small MCP server and TypeScript helpers so the Raycast MCP Extension can call Neocortex tools to save, recall, and delete persistent memory. It follows the same patterns as other memory MCP servers (like the Memory extension and codex-extra-memory) but stores data in Neocortex instead of local databases.
- MCP server for Raycast — exposes the full Mastra-compatible Neocortex tool surface:
neocortex_save_memoryneocortex_recall_memoryneocortex_delete_memoryneocortex_sync_memoryneocortex_insert_documentneocortex_insert_documents_batchneocortex_list_documentsneocortex_get_documentneocortex_delete_documentneocortex_query_memory_contextneocortex_chat_memory_contextneocortex_record_interactionsneocortex_recall_thoughtsneocortex_chat_memoryneocortex_interact_memoryneocortex_recall_memory_masterneocortex_recall_memoriesneocortex_get_ingestion_job
- Shared client — reuses the same
NeocortexMemoryClientand types as the other Neocortex plugins. - Simple bootstrap — one helper (
runNeocortexMcpServerFromEnv) you can point to from the Raycast MCP Extension configuration.
Install into the workspace where you manage your MCP servers:
npm install @neocortex/plugin-raycastBuild the server:
npm run build-
Install the Raycast MCP Extension (from Raycast’s built-in marketplace).
-
Add a new MCP server in the Raycast MCP settings:
- Command:
node - Arguments:
./node_modules/@neocortex/plugin-raycast/dist/index.js
- Environment variables:
TINYHUMANS_API_KEY(required): Neocortex/TinyHuman API key or JWT.TINYHUMANS_BASE_URL(optional): Override Neocortex base URL; defaults to the same staging URL used by other Neocortex plugins.
After saving, Raycast will:
- Launch the MCP server via
node dist/index.js. - Discover the tools:
neocortex_save_memoryneocortex_recall_memoryneocortex_delete_memoryneocortex_sync_memoryneocortex_insert_documentneocortex_insert_documents_batchneocortex_list_documentsneocortex_get_documentneocortex_delete_documentneocortex_query_memory_contextneocortex_chat_memory_contextneocortex_record_interactionsneocortex_recall_thoughtsneocortex_chat_memoryneocortex_interact_memoryneocortex_recall_memory_masterneocortex_recall_memoriesneocortex_get_ingestion_job
- Allow you to
@-mentionthe Neocortex server and invoke its tools in AI Chat, Commands, and Presets.
You can also use the adapter directly from TypeScript/Node without Raycast:
import { RaycastNeocortexMemory } from "@neocortex/plugin-raycast";
const memory = new RaycastNeocortexMemory({
token: process.env.TINYHUMANS_API_KEY!,
baseUrl: process.env.TINYHUMANS_BASE_URL, // optional
defaultNamespace: "my-raycast-workspace",
});
// Save memory
await memory.saveMemory({
key: "preferred_shell",
content: "The user's preferred shell is zsh.",
});
// Recall memory
const recalled = await memory.recallMemory({
query: "What is the user's preferred shell?",
});
console.log("Recalled context:", recalled.context);Internally, this uses the shared NeocortexMemoryClient to call Neocortex’s /v1/memory/* API.
- The MCP server uses
@modelcontextprotocol/sdk’sMcpServer+StdioServerTransportand registers tools usingzodschemas, mirroring the Claude Code and Codex Neocortex plugins. - Credentials never flow through tool inputs; all sensitive configuration is provided via environment variables when the MCP server process is launched.