[Phase 1] Publish to Smithery.ai MCP directory

[izgorodin]

Phase 1 — Quick Wins

Smithery.ai is a community directory of MCP servers. One CLI command to publish.

Update 2026-04-13 (R5 research): The original "no competitors in memory category" assumption is refuted. Pre-publish ranking-signal requirements have been added. See updated sections below. Original body preserved in git history.

Current state of the memory category on Smithery

Verified via Smithery API ?q=memory (see mnemoverse-distribution:research/R5-findings.md §1.1 + data/R5-competitor-presence.csv):

• ~20 memory servers already listed in Smithery under ?q=memory
• 2 verified: mem0ai/mem0-memory-mcp (useCount=385, rank [Phase 1] Publish to Smithery.ai MCP directory #1) and supermemory (verified, rank [Phase 1] List on Glama / MCP.so / PulseMCP directories #2)
• Roughly another 18 unverified memory servers, most created in 2026-03, actively filling the category
• Publishing a minimum-viable smithery.yaml today lands Mnemoverse around rank 20–30 in ?q=memory, not the open space the original ticket implied

Correction propagated to R5-findings.md, audit-analysis.md, and issues-backlog.md on 2026-04-13.

Pre-publish requirements (R5 §2.1)

Smithery uses Reciprocal Rank Fusion (RRF) on the description text as its primary search-match signal, plus a verified badge as a strong secondary boost. Before publishing, we need three things:

1. RRF-optimized description in server.json + smithery.yaml — see [SEO/GEO] Rewrite server.json + smithery.yaml description for Smithery RRF match #19 ([SEO/GEO] Rewrite server.json + smithery.yaml description for Smithery RRF match). Current description has "memory" once and no action verbs; mem0's rank-1 description has "memory" twice plus "semantic", "long-term", "save/search".
2. URL-based publish via smithery mcp publish <https-url> -n @mnemoverse/mcp-memory-server — this is the modern primary path; the stdio-only smithery.yaml path is still supported but not the ranking default.
3. Apply for verified status — process is undocumented, contact Smithery directly after publish.

Publishing without the description rewrite = rank 20-30 = near-zero installs. Do #19 first, this issue second.

Steps (updated)

1. Complete [SEO/GEO] Rewrite server.json + smithery.yaml description for Smithery RRF match #19 — RRF description rewrite in server.json and smithery.yaml
2. Run smithery mcp publish <https-url> -n @mnemoverse/mcp-memory-server (URL-based flow)
3. Verify listing on https://smithery.ai and check ?q=memory rank
4. Request verified status via Smithery contact channel
5. Seed useCount by distributing install links in README badges, docs, and the Mnemoverse website

Test

• [SEO/GEO] Rewrite server.json + smithery.yaml description for Smithery RRF match #19 lands first (description rewritten for RRF, deployed to main + tagged)
• Search ?q=memory on smithery.ai → Mnemoverse appears at rank ≤ 10 within 4 weeks of publish
• Install via Smithery UI → MCP added to Claude / Cursor without error
• memory_write / memory_read works after install
• Verified badge obtained (stretch — undocumented process)

Effort

~30 minutes for the publish itself, plus a 30-minute dependency on #19.

Why (updated)

• Discovery: ~20 memory servers in Smithery's category. We need to rank ≤ 10 to matter.
• Cross-platform: Smithery feeds multiple MCP clients — one publish propagates.
• Competitive gap: Mem0 is rank 1 (verified + high useCount). Graphlit and Cognee are not on Smithery at all (data/R5-competitor-presence.csv). We have a clean shot at rank 2–3 if we execute the RRF rewrite properly.
• Uncontested moat signal: Mnemoverse is already on the Official MCP Registry (v0.3.3, isLatest=true). Mem0, Graphlit, Cognee, and Anthropic's own server-memory reference are all absent from the Official Registry — a real, currently-uncontested moat we should reference in the description copy.

Sources

• mnemoverse-distribution:research/R5-findings.md §1.1 + §2.1 + §5 — full Smithery teardown and pre-publish checklist
• mnemoverse-distribution:research/data/R5-competitor-presence.csv — competitor presence map across 6 registries
• mnemoverse-distribution:research/data/R5-registries.csv — raw registry field data
• Prerequisite issue: [SEO/GEO] Rewrite server.json + smithery.yaml description for Smithery RRF match #19 — [SEO/GEO] Rewrite server.json + smithery.yaml description for Smithery RRF match

Related

• Part of Distribution Sprint — project board feat: single source of truth for all distribution channel configs #6
• R5 SEO/GEO research track — mnemoverse-distribution:research/R5-findings.md

[izgorodin]

Body updated 2026-05-27: file references migrated from mnemoverse-docs/internal/seo-research/ to mnemoverse-distribution/research/ (migration PR: https://github.com/mnemoverse/mnemoverse-docs/pull/50).

[izgorodin]

Status (2026-06-15): server.json is tuned and ready, but verified not yet published — smithery.ai/server/@mnemoverse/mcp-memory-server returns 404 and the Smithery registry API returns "Server not found". Next action: run smithery publish (external/stdio registration). Keeping open until it's actually live.

[izgorodin]

Update 2026-06-25: now listed on Smithery, but the listing routes through Smithery's own hosted proxy rather than our endpoint. Remaining work: re-point it at https://mcp.mnemoverse.com/mcp so connections reach our server directly (needs a Smithery account action). Keeping open for the re-point.