@memoweft/mcp-server

@memoweft/mcp-server

English · 简体中文

An MCP server that exposes a MemoWeft database to an external AI client. It uses MemoWeft's public Core facade and registers exactly eight tools: five reads and three controlled writes.

Install

@memoweft/mcp-server@0.2.3 declares a memoweft peer range of ^0.5.1 || ^0.6.0 || ^0.7.0 || ^1.0.0-rc.1. To use the current stable Core release:

npm install memoweft @memoweft/mcp-server@0.2.3

It resolves cleanly against Core 0.5.1, 0.6, 0.7, the 1.0 RC series, and final 1.x releases with no --legacy-peer-deps. Needs Node 20+.

Tools and data access

Tool	Core method	Kind
memoweft_recall	core.recall	read
memoweft_list_cognitions	core.memory.listCognitions	read
memoweft_list_evidence	core.memory.listEvidence	read
memoweft_list_events	core.memory.listEvents	read
memoweft_graph	core.graph.buildMemoryGraph	read
memoweft_ingest_user_message	core.ingestUserMessage	controlled write
memoweft_ingest_tool_result	core.ingestToolResult	controlled write
memoweft_mute_cognition	core.memory.muteCognition	controlled write

The two ingest tools record one raw evidence item each; they do not update the profile, run consolidation, or change cloud-read authorization. mute_cognition only toggles whether an existing cognition participates in recall; it does not delete it, alter its confidence, or change authorization. Destructive operations, authorization changes, profile updates, observations, and portable import/export are not registered.

Connecting an MCP client grants that client access to database content returned by these read tools, including evidence, cognitions, events, and graph output. Treat the client as a database reader and apply host-level consent and access controls before configuring it. allowCloudRead only qualifies evidence for MemoWeft's built-in cloud-model write prompts; it does not constrain the list, cognition, event, graph, or MCP output surfaces.

memoweft_recall accepts optional contentTypes and explain. With explain, restricted provenance omits its summary, but this is not end-to-end disclosure control for cognitions or other MCP results.

Run

The server talks MCP over stdio. It requires MEMOWEFT_DB_PATH; use :memory: for an ephemeral database.

{
  "mcpServers": {
    "memoweft": {
      "command": "npx",
      "args": ["-y", "@memoweft/mcp-server@0.2"],
      "env": { "MEMOWEFT_DB_PATH": "/absolute/path/to/memoweft.db" }
    }
  }
}

Programmatic use

import { createMemoWeftCore } from 'memoweft';
import { createMcpServer } from '@memoweft/mcp-server';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';

const core = createMemoWeftCore({ dbPath: process.env.MEMOWEFT_DB_PATH! });
const server = createMcpServer(core);
await server.connect(new StdioServerTransport());

Registry

This repository's server.json describes the 0.2.3 server for the official MCP registry; keep it in step with the published npm package and update it when a new version is ready. package.json's mcpName must remain identical to server.json's name.

License

MIT