Agents must check before they claim. That principle becomes release evidence and blocked-state logic.
Proof Surfaces
The operating layer is built from inspectable parts.
Loom, Ground, MCP connectors, and client builds are not a random product shelf. They are proof surfaces for the same path: connect the system, verify the claim, coordinate the agents, then control the workflow.
How to read this page
Products are proof primitives. Trust Layer is the operating layer.
The free and open tools show the discipline underneath the service: grounded claims, agent continuity, constrained MCP access, and evidence-backed decisions. The paid work turns those primitives into one workflow your operator can trust.
Agent work needs memory, ownership, checkpoints, and evidence so progress survives handoffs.
Read-only or constrained MCPs prove the connection before the workflow earns more authority.
The products are proof primitives. The paid work turns them into controlled operating paths.
Flagship proof surfaces
Verify before claiming, then coordinate the work.
The two core open tools behind the operating-layer thesis: Ground checks claims, and Loom gives agent work memory, ownership, and evidence.
Loom
Shared memory, smart task routing, crash recovery. Multi-agent coordination for Claude, Cursor, Codex.
- Memory and coordination for AI agents
- @createsomething/loom-mcp
- Instant setup
Stop hallucinated duplicates. Ground requires verification before claims — no more false positives.
- Code analysis that checks before it claims
- @createsomething/ground-mcp
- Instant setup
Method and control primitives
Framework tools make the delivery philosophy inspectable.
These tools show the rules underneath the service before they become client workflow systems.
Triad Audit
DRY (duplication), Rams (excess), Heidegger (disconnection). Three-level audit with actions: Unify, Remove, Reconnect.
- Subtractive Triad analysis for codebases
- @createsomething/triad-audit
- Instant setup
Interactive MCP server for learning the CREATE SOMETHING approach — authentication, lessons, praxis exercises, and reflection.
- Learn the Subtractive Triad via Claude Code
- @createsomething/learn
- Instant setup
Perceive duplication, excess, and disconnection before dwelling in code. A Gemini extension for learning to see.
- Gemini CLI extension for Subtractive Triad foundations
- @createsomething/seeing
- Instant setup
The framework itself as an MCP server — Resources, Tools, and Prompts. Architecture review, debugging, and design guidance.
- Database, Automation, Judgment as MCP
- Instant setup
MCP-managed workspaces with D1 and R2. Teams interact through agents instead of a UI — structured data, files, and connectivity.
- Agent-native data layer
- Requires Cloudflare setup
Connection wedge MCPs
MCPs prove the connection first.
When the workflow becomes strategic, the same wedge can graduate into approvals, blocked states, and operator briefs.
Notion Sync
Bidirectional sync between a master Issues database and client-specific Notion databases with conflict handling.
- Two-way Notion database sync
- Requires Notion integration
CRUD, backfill, forecast, and conflict detection. Deployable as a Cloudflare Worker with SSE transport for AI clients.
- Shared scheduling with backfill and forecast
- Requires Cloudflare setup
Read-only MCP server giving AI agents secure access to QuickBooks data, synced into Notion databases with OAuth auto-refresh.
- QuickBooks Online to Notion sync
- Requires Intuit + Notion setup
Analyze Webflow sites with a self-improving intelligence layer: touchpoints, SEO, structure, performance, and Designer metadata.
- Site analysis — SEO, structure, performance
- Requires Webflow access
MCP server for the Webflow Marketplace with plagiarism detection (MinHash, LSH, PageRank), marketplace asset analysis, and framework detection.
- Plagiarism detection and marketplace analysis
- Requires Webflow access
Monitor platforms, draft responses, and queue them for human review. Community presence with minimal daily effort.
- Agent-managed community presence
- Instant setup
AI-native social calendar management. Schedule, observe, and optimize LinkedIn content through MCP.
- LinkedIn content scheduling and optimization
- Instant setup
Client workflow evidence
Real systems move from integration work into operating visibility.
Selected builds show how code, runbooks, policy, handoffs, and evidence turn into work an operator can understand.
Three custom MCP servers syncing Gmail emails, YouTube playlist transcripts, and Zoom Clips into Notion — with contact matching, browser automation, and email notifications. Codex setup + policy + runbook included.
- Gmail, YouTube, and Zoom Clips to Notion
- Half Dozen
- Custom project
MCP server for the OUTERFIELDS Premium Content Network — AI-driven component exploration, pattern guidance, deployment, and architecture documentation. Codex setup + policy + runbook included.
- AI-guided platform documentation
- Outerfields
- Custom project
Apply the proof
Need the proof applied to your workflow?
I’ll map the first workflow, identify the safest MCP wedge, and define when the Trust Layer should take over.
Connect Identify the MCP wedge
Name the safest connection point before expanding authority.
Verify Define the proof surface
Decide what evidence proves the workflow worked or stopped correctly.
Control Add the trust layer
Turn the primitive into approval states, blocked states, and operator briefs.