Launched this week

MCP Cursor Guard
The paging bug in your MCP server, fixed with signed cursors
3 followers
The paging bug in your MCP server, fixed with signed cursors
3 followers
MCP pagination fails silently when an agent replays an offset cursor against different filters. Signed, expiring filter-bound cursors fix it: mismatch throws invalid_cursor. Free MIT library on npm plus runnable demo. The kit adds auth, rate limiting, HTTP transport and tests — one payment.



Free
Launch Team

AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
I made this after losing a full afternoon to the same bug.
An MCP server here was returning valid-looking pages of records that answered a different question than the one asked: the agent replayed a plain offset cursor against new filters, and nothing told anyone the page was wrong.
The fix: the cursor now carries a signed, expiring fingerprint of the filters that produced it. Replay against different filters throws invalid_cursor, and every response reports matched / returned / complete so a model can tell a slice from the whole set.
The cursor logic is MIT and free on npm — mcp-cursor-guard — with a runnable demo: npx mcp-cursor-demo-server. The production kit (scoped bearer auth, per-token rate limiting, stdio + Streamable HTTP, 31/31 tests) is at https://legaza.gumroad.com/l/mcp-kit — 79 EUR, one payment, no subscription. Disclosure: I wrote it and I sell it.
This is the kind of bug that scares me most with agents — not a crash, but valid-looking pages answering a different question than the one asked. Binding the cursor to a signed fingerprint of the filters, and having it throw invalid_cursor instead of silently returning the wrong slice, feels like the right fix. The matched / returned / complete counts are a nice touch too — that's exactly what a model needs to know whether it's seen the whole set.