The docs for server card endpoints say:
Each server card includes the standard discovery fields and a tools array that describes the MCP tools the server advertises. Use this to pre-populate tool metadata in MCP clients without making an initialize call to the server.
A client that does that gets an incomplete tool list. On your own docs site:
curl -s https://www.mintlify.com/docs/.well-known/mcp/server-card.json | jq -c '[.tools[].name]'
# ["search_mintlify","query_docs_filesystem_mintlify"]
curl -s -X POST https://www.mintlify.com/docs/mcp \
-H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' \
| sed -n 's/^data: //p' | jq -c '[.result.tools[].name]'
# ["search_mintlify","query_docs_filesystem_mintlify","submit_feedback"]
submit_feedback is served but not advertised. /ai/model-context-protocol#mcp-tools documents it as one of the three tools, so the docs and the endpoint agree with each other — the card is the odd one out.
Not just your site. I run a registry that reads tools/list from public keyless MCP endpoints on a schedule; 51 of the 79 it tracks are hosted Mintlify sites. I checked the card against the live tool list on all 51 today (2026-09-09):
- 50/51 serve
/.well-known/mcp/server-card.json (one 404s)
- 49 of those 50 advertise exactly 2 tools; one advertises 0
- 0 of 50 advertise
submit_feedback
- 51/51 live endpoints serve it
I seeded that server list myself, so 51 is my sample, not a census of your customers.
The changelog looks like it explains the ordering — server cards shipped May 29, 2026, and the submit_feedback tool shipped July 2, 2026, so the card generator may simply predate the third tool.
Two things that might need fixing, and I can't tell from outside which:
- The card generator, if it should be emitting all three tools.
- The sentence recommending the card as a substitute for
initialize, if the card is intentionally a subset — right now it reads as authoritative.
Reproduce fleet-wide:
curl -s https://he110.github.io/mcp-drift-registry/api/registry.json \
| jq -c '[.servers[] | select([.tools[].name]|any(startswith("query_docs_filesystem_")))
| [.tools[].name]|any(.=="submit_feedback")]
| {tenants:length, serving_submit_feedback:(map(select(.))|length)}'
# {"tenants":51,"serving_submit_feedback":51}
Built with an AI coding agent; the measurement is deterministic — fetch, canonicalise, diff — no model in the pipeline.
The docs for server card endpoints say:
A client that does that gets an incomplete tool list. On your own docs site:
submit_feedbackis served but not advertised./ai/model-context-protocol#mcp-toolsdocuments it as one of the three tools, so the docs and the endpoint agree with each other — the card is the odd one out.Not just your site. I run a registry that reads
tools/listfrom public keyless MCP endpoints on a schedule; 51 of the 79 it tracks are hosted Mintlify sites. I checked the card against the live tool list on all 51 today (2026-09-09):/.well-known/mcp/server-card.json(one 404s)submit_feedbackI seeded that server list myself, so 51 is my sample, not a census of your customers.
The changelog looks like it explains the ordering — server cards shipped May 29, 2026, and the
submit_feedbacktool shipped July 2, 2026, so the card generator may simply predate the third tool.Two things that might need fixing, and I can't tell from outside which:
initialize, if the card is intentionally a subset — right now it reads as authoritative.Reproduce fleet-wide:
Built with an AI coding agent; the measurement is deterministic — fetch, canonicalise, diff — no model in the pipeline.