Your identity, not theirs. One file, every AI.
soul.md is an open format for your AI identity — values, voice, skills, taste. A plaintext file you own, version in git, and carry between tools. Manoma is the reference MCP server that makes any LLM read your soul.md.
- User-owned — lives on your machine, in your repo. No cloud, no account, no vendor.
- Portable — works with Claude Desktop, Cursor, Zenflow, and any MCP-aware client.
- Built to outlive models — git-diffable, forkable, versioned. A format, not a walled garden.
See the soul.md format specification.
One line in your Claude Desktop, Cursor, or Zenflow config:
{
"mcpServers": {
"manoma": {
"command": "npx",
"args": ["-y", "manoma-mcp"]
}
}
}Then drop your soul.md at ~/soul.md.
- Web builder — manoma.ai/build — fill the form, download the file.
- Template —
cp templates/founder.md ~/soul.md(also:engineer.md,designer.md,pm.md).
Read — get_context(mode?), get_injection(mode?, message?), get_skill_depth(skill), list_sections().
Write-back — add_decision(...), add_lesson(...), update_now(...).
Resources expose every section as soul://section/<path>, plus soul://full for the whole file.
Full details in mcp/README.md and SPEC.md.
manoma/
├── mcp/ — MCP server (npx manoma-mcp)
├── templates/ — starter soul.md files
├── SPEC.md — format specification
└── README.md
Every AI vendor wants to own your memory. soul.md is the opposite bet: your identity as a plaintext file you control, interchangeable between tools, diff-able in git, outliving any individual model.
The schema is the thing. The runtime is just how it's read.
Contributions, forks, and parallel implementations welcome.
MIT.