HippoDid MCP Server

io.github.snowmanjy由 GitHub 账号 snowmanjy 发布v1.1.0更新于 Oct 24, 2026

Persistent, structured cloud memory for AI agents. BYOK — bring your own AI key.

已验证STDIO仅桌面文件与存储知识与记忆

概览

AI 生成的概览

将 AI 助手接入 HippoDid 云端记忆，使其能跨会话、跨客户端保存、检索和同步结构化记忆。

可以做什么
  • 创建并列出用于划分记忆命名空间的角色
  • 通过显著性/去重流程添加记忆或直接写入，并用语义加关键词的混合检索查找记忆
  • 将本地文件同步到云端、注册监视路径、强制同步并查看同步状态
  • 导入文档以抽取记忆，并将某个角色的全部记忆导出为 Markdown
  • 配置并测试用于抽取的自带密钥 AI 提供商
适用场景
适合希望助手上下文在会话之间、在不同 MCP 客户端或不同 AI 提供商之间保持延续的场景，也适合把项目文件和文档转化为可召回的结构化记忆。
运行要求
以 Java JAR 通过 stdio 在本地运行，需要 Java 运行时（从源码构建需 Java 21+）。需要 HippoDid 账号，并在 HIPPODID_API_KEY 中提供 API 密钥；HIPPODID_CHARACTER_ID 与 HIPPODID_BASE_URL 为可选项。需要访问 HippoDid API 的网络，BYOK 抽取还需自备 OpenAI、Anthropic 或 Gemini 密钥。
安装前请注意
HIPPODID_API_KEY 是凭据，不应写入共享配置。同步的文件、导入的文档和记忆都会上传到 HippoDid 云端，后台监视可能自动发送文件内容。部分工具会写入、导入或导出数据，直接写入与文档导入仅限付费套餐。自带密钥的提供商密钥会保存在租户配置中。

安装

在 SourceWeft 中

  1. 打开 控制台中的 HippoDid MCP Server，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only，通过 STDIO。 STDIO 服务会启动本地进程，因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

[HippoDid MCP Server — persistent cloud memory for AI agents]

HippoDid MCP Server
Persistent, portable memory for AI agents. Your AI forgets everything between sessions. HippoDid remembers.

HippoDid MCP Server

MCP (Model Context Protocol) server for HippoDid — connects Claude Code, Claude Desktop, ChatGPT, Gemini CLI, Codex CLI, Cursor, and any MCP-compatible client to persistent AI character memory.

What is HippoDid?

HippoDid is a cloud memory infrastructure for AI agents. It stores structured memories (preferences, decisions, skills, relationships) in a searchable database and retrieves them across sessions, tools, and AI providers. Write a memory in Claude Code, recall it in ChatGPT. Switch from Cursor to Gemini — your agent's memory travels with it.

Key features:

  • Cross-platform memory — same memories accessible from any MCP client or ChatGPT via OAuth
  • Structured extraction — raw text is automatically parsed into categorized facts with salience scores
  • Temporal decay — memories fade naturally by category (skills are evergreen, events fade in 14 days)
  • Character scoping — each agent/project gets its own memory namespace
  • BYOK AI — bring your own OpenAI, Anthropic, or Gemini key for extraction

Quick Start

1. Get an API Key

Sign up at hippodid.com and create an API key from your dashboard.

2. Download the JAR

Download the latest release from GitHub Releases.

3. Configure Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

json
{  "mcpServers": {    "hippodid": {      "command": "java",      "args": ["-jar", "/path/to/hippodid-mcp-server-1.1.0.jar"],      "env": {        "HIPPODID_API_KEY": "hd_key_your_key_here",        "HIPPODID_CHARACTER_ID": "your-character-uuid"      }    }  }}

4. Start Using

Ask Claude to remember things, search memories, sync files, and more.

MCP Tools

ToolDescription
create_characterCreate a new character to store memories for
list_charactersList all characters accessible to you
add_memoryAdd a memory via the AUDN pipeline (rule-based salience + dedup)
add_memory_directWrite a memory directly, bypassing AUDN pipeline (Starter+ tier)
search_memoriesHybrid semantic + keyword search across memories
sync_fileSync a local file to the HippoDid cloud
import_documentImport a document and extract memories (Starter+ tier)
list_synced_filesList files synced to the cloud for a character
get_sync_statusGet sync status summary for a character
export_characterExport all memories for a character as Markdown
add_watch_pathSync a file and register the path for background tracking
list_watch_pathsList all watched file paths in this session
force_syncForce an immediate sync of all watched paths
configure_aiConfigure tenant BYOK AI providers
test_ai_configTest connectivity of saved AI provider configuration

Environment Variables

VariableRequiredDefaultDescription
HIPPODID_API_KEYYes—Your HippoDid API key
HIPPODID_CHARACTER_IDNo—Default character UUID for background sync
HIPPODID_BASE_URLNohttps://api.hippodid.comAPI base URL
MCP_SYNC_INTERVALNo300Background sync interval in seconds
MCP_AUTO_CAPTURENofalseDisable background watcher (AI handles sync)
MCP_AUTO_RECALLNofalseDisable background hydration (AI handles download)
MCP_RECALL_CACHE_TTLNo120Search result cache TTL in seconds

Architecture

Claude Desktop → stdio (JSON-RPC) → McpServerRunner → HippoDidClient → HTTP → HippoDid REST API

The MCP server uses the hippodid-spring-boot-starter to communicate with the HippoDid REST API over HTTP. All authentication and tenant isolation is handled server-side via Bearer tokens.

Building from Source

Requires Java 21+ and Maven 3.8+.

bash
mvn clean package -DskipTestsjava -jar target/hippodid-mcp-server-1.1.0.jar

Versioning

The MCP server version matches the hippodid-spring-boot-starter version. Both must be at the same version for compatibility.

License

Apache License 2.0 — see LICENSE.

来源：README.md，提交 aab1720

工具

0
SourceWeft 尚未在沙箱中启动过该软件包，因此其工具未知。

版本历史

1
  1. v1.1.0最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手；SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头，或保持端点能够响应探测，以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list，或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分