Skip to content

README.md

Latest commit

 

History

History
92 lines (77 loc) · 4.23 KB

README.md

File metadata and controls

92 lines (77 loc) · 4.23 KB
Raw
name SekaiSync
purpose Select the correct access protocol for a SekaiSync agent
protocols
id label command transport endpoint when directory
mcp-stdio
MCP stdio (local process)
python -m sekaisync serve-mcp
stdio
agent can spawn a local subprocess (desktop app, CLI, IDE)
mcp-stdio/
id label command transport endpoint when directory
mcp-http
MCP Streamable HTTP
python -m sekaisync serve-http
streamable-http
/mcp
agent can only reach a public HTTPS URL (web / cloud agents)
mcp-http/
id label command transport endpoint when directory
rest-openapi
REST + OpenAPI
python -m sekaisync serve-http
http
/api/v1/* and /openapi.json
agent only supports OpenAPI Actions or a plain JSON HTTP API
rest-openapi/

SekaiSync Agent Access (Protocol Selection)

English | 中文

SekaiSync exposes exactly three protocols to Agents. Before integrating, an Agent reads this file, picks the protocol it can connect to, then takes the matching config template from the corresponding directory. Do not build interfaces per Agent brand: the same protocol works for every Agent that supports it.

The three protocols

Protocol Transport Entry When to use
mcp-stdio stdio (local process) python -m sekaisync serve-mcp Agents that can spawn a local subprocess (desktop app / CLI / IDE)
mcp-http Streamable HTTP /mcp of serve-http Agents that can only reach a public HTTPS URL (web / cloud Agents)
rest-openapi HTTP JSON /api/v1/* and /openapi.json of serve-http Agents that only support OpenAPI Actions or plain REST (legacy fallback)

Selection rules (machine-readable)

  1. Can spawn a local process → use mcp-stdio, pick the platform config under mcp-stdio/.
  2. Can only fill in an HTTPS URL → use mcp-http, see mcp-http/.
  3. Only OpenAPI/REST available → use rest-openapi/ (fewest capabilities, not the primary path).

Agent → protocol mapping (machine-readable)

agent protocol config / notes
Codex CLI / ChatGPT desktop / IDE extension mcp-stdio mcp-stdio/codex.config.toml
Claude Code / Claude Desktop mcp-stdio mcp-stdio/claude-code.mcp.json + claude-desktop.json
Cursor mcp-stdio mcp-stdio/cursor.mcp.json
Grok Build mcp-stdio mcp-stdio/grok.config.toml + grok.mcp.json
Hermes Agent mcp-stdio mcp-stdio/hermes.yaml
OpenClaw mcp-stdio mcp-stdio/openclaw.json
AstrBot mcp-stdio mcp-stdio/astrbot.json
OpenCode mcp-stdio / mcp-http mcp-stdio/opencode.json (local); remote via mcp-http/README.md
WorkBuddy (Tencent Cloud Code Assistant) mcp-stdio mcp-stdio/workbuddy-mcp.json
TRAE (ByteDance) mcp-stdio / mcp-http mcp-stdio/trae.json (stdio); remote via Streamable HTTP
ZCode mcp-stdio mcp-stdio/zcode-config.json (.zcode/config.json)
DeepSeek Harness plugin deepseek-harness/ (Cordis plugin bridging MCP)
ChatGPT Work / ChatGPT Web mcp-http mcp-http/chatgpt-work.md + developer-mode.md
ChatGPT Actions (legacy fallback) rest-openapi rest-openapi/openapi.json

Behaviour layer (protocol-independent)

How to use SekaiSync after connecting does not change with the protocol and lives in two places:

  • SKILL.md: the skill loaded only when a Project Sekai question is hit (workflow + output contract).
  • rules/: persistent hard rules copied into a project under the platform-required file names: AGENTS.md (Codex / Grok), CLAUDE.md (Claude Code), SekaiSync.mdc (Cursor), SekaiSync.prompt.md (OpenClaw).

Directory layout

agents/
  README.md            this file: machine-readable access selection
  SKILL.md             shared skill (single source)
  mcp-stdio/           protocol 1: local-process MCP config templates
  mcp-http/            protocol 2: Streamable HTTP (HTTPS /mcp)
  rest-openapi/        protocol 3: REST + OpenAPI
  rules/               persistent behaviour rules (per platform file name)

All example paths (C:\path\to\python.exe, C:\path\to\sekaisync) are placeholders — replace them with your real paths when integrating.