Glyph Hold MCP
MCP server that connects Codex and other MCP clients to a running Glyph Hold instance via its HTTP API, enabling memory and secret management.
README
Glyph Hold MCP
Local stdio MCP server for connecting Codex and other MCP clients to Glyph Hold.
This repo is intentionally separate from Glyph Hold. It does not access the
SQLite database directly. It talks to a running Glyph Hold instance through the
public /api/v1 HTTP API.
Requirements
- Python 3.12+
- A running Glyph Hold instance
- A Glyph Hold API key created from the dashboard
Example URL:
GLYPHHOLD_URL=https://glyphhold.example.com
Local Setup
Choose a place on your machine where you keep local tool repos. Clone this repo there:
cd ~/coding_projects
git clone git@github.com:Dosk3n/glyphhold-mcp.git
cd glyphhold-mcp
Create the local Python environment:
python3.12 -m venv .venv
. .venv/bin/activate
pip install -e ".[dev]"
Create the local environment file:
cp .env.example .env
Edit .env:
nano .env
Set:
GLYPHHOLD_URL=https://your-glyphhold-host.example.com
GLYPHHOLD_API_KEY=gh_live_xxxxxxxxxxxxxxxxx
Create GLYPHHOLD_API_KEY from the Glyph Hold dashboard.
Useful scopes:
memories:readmemories:writesecrets:writesecrets:reveal
Only grant secrets:reveal if you want the MCP client to be able to reveal
secret values after an explicit user request.
Codex CLI Config
Find the full path to the cloned repo:
pwd
If pwd prints:
/home/you/coding_projects/glyphhold-mcp
then add this to ~/.codex/config.toml:
[mcp_servers.glyphhold]
command = "/home/you/coding_projects/glyphhold-mcp/.venv/bin/python"
args = ["-m", "glyphhold_mcp.server"]
cwd = "/home/you/coding_projects/glyphhold-mcp"
Use your real path from pwd. The important parts are:
command = <repo path>/.venv/bin/python
cwd = <repo path>
You do not need to put the API key in Codex config. The MCP server loads
GLYPHHOLD_URL and GLYPHHOLD_API_KEY from the .env file in the cloned repo.
Start Codex from any project as normal.
Inside Codex, run:
/mcp
You should see the glyphhold MCP server connected.
TLS Certificate Errors
If a tool returns an error like:
[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed
Python does not trust the certificate chain for your Glyph Hold URL.
Best fix: point the MCP server at the CA bundle that signs your internal certificate:
GLYPHHOLD_CA_BUNDLE=/path/to/ca-bundle.pem
For a trusted private test deployment, you can disable verification in .env:
GLYPHHOLD_VERIFY_SSL=false
Do not use GLYPHHOLD_VERIFY_SSL=false for an internet-facing deployment.
Updating
To update the local MCP server later:
cd ~/coding_projects/glyphhold-mcp
git pull
. .venv/bin/activate
pip install -e ".[dev]"
Tools
Health and categories:
glyphhold_healthlist_categories
Memory tools:
list_memoriesget_memorysearch_memoriesprefetch_memoriesfind_similar_memoriesprepare_memory_writecreate_memoryupdate_memoryupdate_memory_confidencearchive_memorysupersede_memorylist_memory_revisionsrestore_memory_revisiondelete_memory
Secret tools:
search_secretsget_secret_metadatacreate_secretupdate_secretdelete_secretreveal_secretreveal_secret_env
Secret values are only returned by reveal_secret and reveal_secret_env.
Permanent deletes require an exact confirmation value.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。