Daily scanner tracks gap between MCP registry declarations and actual server responses
art2meta · reddit · 2026-10-13
The author of lowdown-proxy added a daily GitHub Actions scanner probing 2,000 MCP registry entries that declare streamable-http remotes, recording the gap between what the registry declares and what actually responds—snapshots, not uptime.
Findings
- 14 entries unprobeable on Oct 11: 12 delisted, 2 with no declared remote
- One host serving 130 endpoints went 32→129→40 reachable across three days, motivating separate logging of first attempts vs. retries
Recent changes: auto-retries for transient failures, per-entry gap reasons, npx lowdown-proxy zero-arg startup, registry API paging fallback, a lowdown-mcp-lookup skill on clawhub, and CSV mirroring to Hugging Face. Data CC BY 4.0, code MIT.
Limitations: no stdio-only coverage, single network location, and no confirmed external agent usage of the lookup API yet. The author declined paid directory listings and asks the community which history (reachability, schema changes, agent-reported outcomes) would actually help server selection.
More from coding & agent
- Anthropic Post: Claude Rewrote Pipeline With 20x Less Memory and 4x Speed — teortaxesTex · 2026-10-14
- Cursor now ships real Supabase backend changes via MCP, no SQL needed — XFreeze · 2026-10-14
- Developer: superhuman agents finished the last 20%, so vibe coding is dead — josh_bickett · 2026-10-14
- MIRA-AC trains research agents on proxy feedback, lifting symbolic regression from 50 to 63 solved — anirudhg9119 · 2026-10-14
- MIRA-AC behavior shift: hypothesis-distinguishing tests up 16.8% to 28%, stalled tweaks halved — anirudhg9119 · 2026-10-14
- MIRA-AC RL details: stitching decisions into one logical episode, updating only the meta-reasoner — anirudhg9119 · 2026-10-14