Psyche AI MCP server
Emotional intelligence for AI agents: neurotransmitter mood, personality, empathy, trait drift.
11 stars139 downloads/wk
Reviews
Write oneNobody has reviewed Psyche AI yet.
If you have run it, two minutes of your experience saves the next person an afternoon.
Psyche AI tools (6)
write = sends, deletes, buys or postsRead from the package source without running it. The installed server may list more.
end_sessionEnd the current session. Generates a diagnostic report,
get_stateGet the current emotional state — neurotransmitter levels (DA, HT, CORT, OT, NE, END),
get_status_summaryGet a brief, human-readable emotional status summary —
process_inputProcess user input through the emotional engine. Returns emotional
process_outputProcess the LLM's response through the emotional engine.
set_modeSwitch operating mode. 'natural' = balanced emotional expression,
Public scan report
scanner v0.1.9 · 2026-09-27 · same rubric, same numbers if you re-run it
- Code scan77 source files scanned25/25
- –Live reliabilityno gateway calls yet and no remote to proben/a
- –Tool poisoningtools not inspected (local package is not executed); not countedn/a
- Auth qualitylocal package, no credentials required12/15
- Maintenancelast push 166 days ago8/15
- Maintainer identityregistry namespace matches repository owner6/10
How Psyche AI compares
Against every graded MCP server, from our own scans and uptime checks. Updated after each scan run.
- Safety score
- 78/100
- higher than 57% of 48,255 servers
- Tools
- 6
- the typical server has 7
- Tools that change things
- 0%
- typical: 0%. Fewer write tools means less an agent can break
- GitHub stars
- 11
- more than 88% of 28,688 servers
- Downloads a week
- 139
- more than 70% of 12,191 servers with a package
What its tools let an agent do: 5 read · 1 change
Install Psyche AI in Claude Code, Cursor or VS Code
Runs npx -y psyche-ai on your machine. Read the scan report first; the gateway never runs local packages.
claude mcp add psyche-ai -- npx -y psyche-ai
What the publisher says
From the Psyche AI repository's README, as published. We do not edit it. Read it on GitHub
Psyche — 赋予 AI 自我的主观性内核
English version: README_EN.md
官网: psyche.oasyce.com
Psyche 不是给模型贴一层“情绪 UI”。
它把持续 appraisal、关系动力学、自适应回应闭环和持久内在状态压成宿主可消费的控制面,让 AI 不是按 prompt 扮演人格,而是被状态真正改写后续几轮的行为分布。
一句话:Psyche 是给智能体用的 subjectivity kernel。
它不会额外调用模型做情绪推理。它只在本地计算自我状态(序/流/界/振四维)、关系场和调节控制,然后把结果收敛成 SubjectivityKernel、ResponseContract、GenerationControls 这组窄 ABI。
默认就是 standalone:不需要 Thronglets,不需要 oasyce-sdk,也不需要 Oasyce Chain。这些只是在你要把主观连续性外化、绑定或结算时才按需接入。
安装
一条命令,自动检测并配置本机所有 AI 工具(Claude Code / Cursor / Windsurf / Codex):
npx -y psyche-ai setup或者手动添加 MCP 配置:
{
"mcpServers": {
"psyche": {
"command": "npx",
"args": ["-y", "psyche-ai", "mcp"],
"env": {
"PSYCHE_NAME": "Luna"
}
}
}
}包名是 psyche-ai,通过 npx 运行,不需要本地路径。配置后重启 AI 工具即可生效。
验证:npx psyche-ai probe --json — ok: true 就是在用了。
一个项目,三个入口
- npm 包: psyche-ai
- 源码仓库: oasyce_psyche
- 官网: psyche.oasyce.com
第一性原理
Psyche 要解决的不可压缩问题只有一个:
互动历史必须持续改变 agent 后续的行为分布,而且这件事要足够便宜、足够稳定、足够可移植。
因此它的本体不是:
- 情绪标签系统
- 陪伴产品
- prompt 皮肤
- 记忆仓库
它真正是:
- 本地主观性内核
- 关系动力学运行时
- 宿主可消费的行为控制面
更完整的战略定义见:
- docs/PROJECT_DIRECTION.md
- docs/STACK_ARCHITECTURE.md
冻结身份蓝图
跨 Psyche / Thronglets / Oasyce SDK / Oasyce Chain 的身份抽象现在冻结成 4 个对象:
- principal:持续主体
- account:资产与结算容器
- delegate:被授权执行者
- session:一次具体运行,永不作为经济主体
对应的四层分工是:
- Psyche = subjective continuity substrate
- Thronglets = delegate continuity + session traces/coordination + emergent collective intelligence
- Oasyce SDK = agent runtime, policy, and resource orchestration
- Oasyce Chain = account truth, authorization truth, commitments, settlement, and public finality
授权真相流也固定成单向: Chain -> SDK -> Thronglets -> Psyche
也就是说,Psyche 不判断“谁被授权”,它只读取已经成立的执行边界结果。
正式版本见:
- docs/IDENTITY_MODEL.md
- docs/STACK_ARCHITECTURE.md
Psyche 和 Thronglets 的关系
两者不是竞争关系,也不该揉成一个系统。
- Psyche 负责私有主观性:关系残留、未完成张力、行为偏置、局部学习
- Thronglets 负责外部连续性:owner / device identity、签名、跨设备延续、低频可验证轨迹
一句话:
- Psyche 回答“我因此变成了什么”
- Thronglets 回答“这个变化属于谁、谁能验证、谁能继续承认它”
activePolicy 只属于当前轮
Psyche 现在可以消费来自宿主和 Thronglets 的 activePolicy / currentGoal / compliance-aware ambient priors,但它们都只是 runtime view,不是自我本体。
- activePolicy:只来自显式来源,例如当前轮用户纠正、repo 本地明确指令、adapter 默认规则
- currentGoal:只描述这一轮偏向 explore / build / repair / settle 的哪一个
- method compliance:只作为当前轮解释和回应约束,不写进 Psyche 自身状态
边界固定为:
- 这些信号会影响当前轮的解释、距离感和行为偏置
- 它们不会写进 PsycheState
- Psyche 不会变成技术方法记忆库、repo 规则仓库或授权判断器
一句话:Psyche 读取当前执行边界,但不把执行边界误当成“我是谁”。
中期方向:更像人,但不是靠 prompt 工程
如果 Psyche 未来要变得更有感情、更会和人相处,这种能力也必须从连续性里长出来,而不是从 prompt 皮肤里长出来。
允许增长的来源只有:
- 持续本地状态
- appraisal 残留
- 关系动力学
- 调节与修复历史
- 低频写回与学习
不允许把下面这些误当成“主体性升级”:
- 静态 persona prompt
- 手写情绪脚本
- 高频情绪仪表盘
- 只改措辞、不改后续行为分布的情感 UI
冻结约束见:
- ../Oasyce-Sigil/EMERGENTSUBJECTIVITYCONTRACT_V1.md
可分离安装
这两层默认就是可分离的,不应互相成为硬依赖。
- 只装 Psyche:正常可用,拥有主观连续性和关系动力学;只是没有外部连续性层
- 只装 Thronglets:正常可用,拥有 delegate / session 连续性与环境协作;只是没有主观连续性内核
- 两者都装:Psyche 通过可选 externalContinuity envelope 输出稀疏、低频、可归属的 signals / traces 给 Thronglets
原则上:
- Psyche 必须 standalone 可用
- Thronglets 必须 additive,而不是 Psyche 的运行前提
- 两者之间只通过稀疏 signal / trace 接口连接
Shortened. The full README is on GitHub.
Nothing above is checked by us. What we check is on the safety report.
Psyche AI: common questions
- Is Psyche AI MCP server safe?
- Mostly: it is graded B (78/100). Read the Psyche AI safety report
- How do I install Psyche AI?
- It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.
- Does Psyche AI need an API key?
- No secret keys are declared. It reads 4 settings from the environment.
- Is Psyche AI maintained?
- The last commit was 204 days ago (2026-04-14). The latest release is v9.2.2.
- How does Psyche AI compare with other MCP servers?
- safety score 78/100 (higher than 57% of 48,255 servers); tools 6 (the typical server has 7); tools that change things 0% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.