Tuğberk AkbulutGet in touch
Back to work

openGlad

2025–2026

An MCP server that talks a founder out of building things nobody wants.

TypeScriptCloudflare WorkersModel Context ProtocolReddit APIZod
Source Live
openGlad logo

An open-source MCP server, live at openglad.tuguberk.dev/mcp, positioned as a loss-prevention friction engine for founders. It exposes 13 tools directly inside Claude Desktop, Cursor, Windsurf, and Le Chat, so the reality check happens where the building already happens, before a line of code gets written.

Architecture

AI Client Claude · Cursor Windsurf MCP openGlad Worker Cloudflare Edge 13 tools Reddit · 11 subs HN · Algolia GitHub Search Polymarket parallel fetch, cached 1hr at the edge

All four sources are free, public APIs with no key required. Results are deduplicated across sources with Jaccard N-gram similarity, capped per author to prevent single-voice dominance, and ranked by an engagement score that weights freshness, score, and activity together.

Tool inventory

Friction engine (loss prevention): run_the_bet (mega-pipeline combining pattern scan, loss simulation, and revenue gate; the recommended starting point for a new idea), pattern_scan (behavioral risk detection: overbuilding drift, monetization avoidance, prestige bias), loss_simulation (three-scenario failure predictions with quantified expected loss), revenue_gate (locks building until a monetization strategy is confirmed), compare_ideas (parallel analysis of two or three ideas with a single ranked verdict).

Market intelligence: analyze_market_trends (overcrowding and late-entry risk detection), scan_reddit_trends (broad sentiment scan with 6 to 12 month predictions).

Startup diagnostics: analyze_startup (triage router), analyze_execution_stability, analyze_revenue_health, analyze_burnout_risk, analyze_distribution_discipline, generate_full_diagnosis (a full scan across every diagnostic dimension at once).

How a run works

A user says “I want to build an AI resume builder.” The client calls run_the_bet. The worker fetches all four sources in parallel, deduplicates and ranks the results, runs the pattern scan for behavioral risk, simulates three failure scenarios grounded in the real market signal it just pulled, and locks the revenue gate until monetization is proven. What comes back is a blunt reality check: blind spots, failure modes, and whether the idea is allowed to be built at all, not a motivational pitch.