# Ogham MCP — Shared Persistent Agent Memory

> Ogham MCP provides persistent, searchable shared memory for AI coding agents across clients, with a CLI wizard (`ogham init`) and stdio/SSE transports.

## Install

Copy the content below into your project:

## Quick Use

```bash
# recommended wizard
uvx --from ogham-mcp ogham init

# if you use Neon/self-hosted Postgres
uvx --from 'ogham-mcp[postgres]' ogham init

# add to Claude Code (manual)
claude mcp add ogham -- uvx ogham-mcp
```

## Intro

Ogham MCP provides persistent, searchable shared memory for AI coding agents across clients, with a CLI wizard (`ogham init`) and stdio/SSE transports.

- **Best for:** Long-lived memory for multi-client/multi-agent coding workflows
- **Works with:** Python + uv/uvx; Supabase or Postgres; Claude Code / Cursor / OpenCode configs (per README)
- **Setup time:** 15–40 minutes

## Practical Notes

- GitHub: 97 stars · 19 forks; pushed 2026-05-05 (verified via GitHub API).
- README quick start uses `uvx --from ogham-mcp ogham init` to write client configs and migrate schemas.
- README documents SSE mode with `ogham serve --transport sse --port 8742` for multi-agent sharing.

## Main

How to use memory without turning it into noise:

1. Decide what is “stable knowledge” (architecture decisions, conventions, runbooks) vs “ephemeral state”.
2. Only persist stable knowledge, and enforce a simple schema (component, decision, rationale, date).
3. When a task finishes, have the agent write a short memory entry, then validate retrieval by asking a new agent to solve a similar task.

If you have multiple agents running, prefer SSE mode so they share one server instance and database pool.

### FAQ

**Q: Do I need Supabase?**
A: No—README supports Supabase or Postgres (Neon/self-hosted).

**Q: How do I connect different clients?**
A: Run the init wizard; it prints/writes client config snippets for common MCP clients.

**Q: What transport should I use?**
A: stdio is simplest; use SSE when multiple agents should share one running server.

## Source & Thanks

> Source: https://github.com/ogham-mcp/ogham-mcp
> License: MIT
> GitHub stars: 97 · forks: 19

---

<!-- ZH -->

## 快速使用

```bash
# 推荐：向导式初始化
uvx --from ogham-mcp ogham init

# 如果使用 Neon/自建 Postgres
uvx --from 'ogham-mcp[postgres]' ogham init

# 手动添加到 Claude Code
claude mcp add ogham -- uvx ogham-mcp
```

## 简介

Ogham MCP 为 AI coding agent 提供持久、可检索的共享记忆：支持 `ogham init` 向导配置数据库与 embedding，并提供 stdio/SSE 两种传输模式，跨客户端共享同一份上下文。

- **适合谁：** 多客户端/多 agent 的长期记忆与上下文共享
- **可搭配：** Python + uv/uvx；Supabase 或 Postgres；可生成 Claude Code/Cursor/OpenCode 配置（见 README）
- **准备时间：** 15–40 分钟

## 实战建议

- GitHub：97 stars · 19 forks；最近更新 2026-05-05（GitHub API 验证）。
- README quick start 用 `uvx --from ogham-mcp ogham init` 写入客户端配置并完成 schema 迁移。
- README 给出 SSE 模式：`ogham serve --transport sse --port 8742`，适合多 agent 共用一套服务端。

## 主要内容

如何避免“记忆变噪音”：

1. 先区分“稳定知识”（架构决策、规范、runbook）与“临时状态”。
2. 只把稳定知识写入，并用简单结构约束（模块/决策/原因/日期）。
3. 每次任务结束写一条短记忆，再用另一个 agent 做一次检索回放，确认能找回。

多 agent 并行时，优先用 SSE 模式，共用一个服务端实例与数据库连接池。

### FAQ

**必须用 Supabase 吗？**
答：不必须。README 支持 Supabase 或 Postgres（Neon/自建）。

**不同客户端怎么接？**
答：跑 init 向导，它会输出/写入常见 MCP 客户端的配置片段。

**传输选哪种？**
答：stdio 最简单；多 agent 共用时选 SSE 更合适。

## 来源与感谢

> Source: https://github.com/ogham-mcp/ogham-mcp
> License: MIT
> GitHub stars: 97 · forks: 19


---
Source: https://tokrepo.com/en/workflows/ogham-mcp-shared-persistent-agent-memory
Author: MCP Hub