jpicklyk/task-orchestrator

docker pull ghcr.io/jpicklyk/task-orchestrator:latest
Recommended: run a persistent HTTP server (MCP_TRANSPORT=http, -p 127.0.0.1:3001:3001) and register in .mcp.json as {"type":"http","url":"http://localhost:3001/mcp"}
Simpler: STDIO shape using docker run --rm -i -v mcp-task-data:/app/data ghcr.io/jpicklyk/task-orchestrator:latest

Usage

• •Define YAML work_item_schemas (notes, roles, required, guidance, skill) in .taskorchestrator/config.yaml
• •Agents create work trees, fill notes, advance items; the server blocks transitions until gates pass
• •Enable actor_authentication to require actor claims on every write; recover full state with get_context(since=...)

Key features

• ✓Server-side enforcement: advance_item errors until required notes are filled; dependencies validated before transitions
• ✓Composable traits (e.g. needs-security-review) add cross-cutting note requirements to any schema
• ✓Persistent WorkItem graph (up to 4 levels) with typed dependency edges, cascading completion and atomic work-tree creation
• ✓Actor attribution with pluggable verification (JWKS), full-text search with relevance ranking, and session continuity
• ✓Kotlin + SQLite/Exposed with FTS5, Flyway migrations; optional REST API with SSE and bearer/JWKS auth

Best for

Multi-agent teams that need structural workflow enforcement and accountability the model can't provide.

Caveats

• ⚠HTTP + REST without auth is only safe bound to loopback (127.0.0.1) — never expose on a wider interface
• ⚠Without schemas, all 14 tools run in schema-free mode — add schemas to enable gates
• ⚠The Claude Code plugin is optional; the MCP server works standalone