mcp-quality
Portable Agent Plugin with skills for designing, reviewing, and scaffolding MCP servers — especially knowledge-graph backends with a shared HTTP API for reports and human UIs.
This pack is guardrails and workflow, not a running MCP server or graph database. Agents use the skills to build those in your project.
Install
- Open this repo as a Cursor / Agent Plugin project, or install via Cursor Plugins from the package root (where
plugin.jsonlives). - Skills load from
skills/*/SKILL.mdautomatically.
Skills
| Skill | When to use |
|---|---|
| mcp-quality-server | Any MCP design or review: envelopes, gates, playbooks, discovery_path, discovery attestation, shared domain API |
| mcp-knowledge-graph | Graph/ontology tool design (after quality) |
| mcp-kg-builder | “Build/maintain a KG MCP for my use case” — interview → confirm model → scaffold core + MCP + API |
There is no fourth API-only skill: shared HTTP API guidance lives in the quality skill and builder.
Authority
- Agents: follow the skills. When the long essay and skills disagree, skills win.
- Humans: read
docs/mcp-standards.mdfor the narrative quality bar (product appendix at the end is history, not portable requirements). Case study:docs/mcp-has-no-quality-standards-we-wrote-ours-so-agents-can-run-our-website.md. - Sync the essay when you change the contract in skills.
Frictionless local access (default)
For non-technical builds, scaffolded servers should:
- Bootstrap a work session before serious work and before minting UI links
- Issue 2-hour signed UI links tied to
work_session_id(remint in chat on failure; warn not to forward) - After first open, use a same-day (~8–12h) browser session
- Keep UI read-oriented; high-blast writes stay on the agent/MCP path with discovery attestation
Out of scope for this pack
- No bundled graph DB or live MCP process
- No website-v3 / product CMS paths in skills
- No Cursor-only
rules//commands/packaged in the plugin