MCPorchestrationpublic demo A single layer between your projects, your connectors, your sources and your AI agents.
⚠ Public demo. This instance serves fictional sample projects only, kept in memory (ephemeral — reset per instance). No real, private or personal data. The production ProjectCore with real projects is a separate, authenticated server.
ProjectCore is an orchestration layer for ChatGPT, Claude, Claude Code and Codex. It coordinates your projects and your MCP connectors: registry, routing (« where to look? »), method (decisions, provenance, versions), Context Packs for handoff, and a Latest Artifact Resolver module to find the latest good version of a deliverable.
This server has two parts. The orchestration brain (registry, routing, method, handoff
structure) is stateless. Since phase 2 (done) it also provides a
durable project memory (the pos_* tools): decisions (Decision Ledger,
append-only), actions/deadlines, verified sources (Source Truth), artifact versions and Context Packs. In
this public demo they are kept in an in-memory store seeded with fictional sample
projects (ephemeral, reset per instance) — never a real or personal record. It still does
not retrieve files itself: real file/project data comes from your connected
sources. See the projectos_modules tool for each module's maturity.
Orchestration brain (read-only): projectos_apercu · projectos_registre ·
projectos_modules · projectos_router · projectos_context ·
projectos_recherche · projectos_artifact · projectos_methode.
Durable project memory: pos_project_* · pos_decision_* ·
pos_action_* · pos_source_* · pos_artifact_* ·
pos_context_* · pos_audit_log.
Invent nothing; keep provenance (Source Truth); target the CURRENT version; to find a deliverable, apply the method then surface the real file through a connected source.
/api/mcp — Streamable HTTP transport (stateless), no authentication. The durable project
memory (pos_* tools) is a shared store with no per-user isolation — see Privacy.