Recipe-Inhalt ist auf Englisch. Englisches Original lesen →
← Alle Recipes
Phase 9 · Distribution·9 steps

MCP marketplaces, where to submit, what each one auto-pulls

mcp.so, MCPize, awesome-mcp-servers, Smithery, Glama. Submission process, what each directory auto-pulls from your README, how to spot fake-traffic ones.

9 steps0%
Du liest ohne Account. Mit Login speichern wir Step-Fortschritt + Notes.

MCP marketplaces, where to submit, what each one auto-pulls

After your server is on npm + GitHub, the question is: how do users find it? The MCP ecosystem has ~7 directories that matter, plus a long tail of low-traffic ones. This recipe is the priority order, the submission flow per directory, and how to recognize the ones that aren't worth your time.

Schritt 1: Priority order (Q2 2026)

| Directory | Approx. server count | Submission | Auto-pulls from | |---|---|---|---| | Official MCP Server Registry (registry.modelcontextprotocol.io) | Authoritative, downstream sync source | mcp-publisher publish (CLI + server.json) | npm + GitHub via package manifest | | Glama (glama.ai) | ~21k indexed | Auto-indexes from GitHub topics | GitHub topics + README | | mcp.so | ~19k indexed | Form + GitHub URL | README + package.json | | PulseMCP (pulsemcp.com) | ~12k indexed | Auto-syncs from Registry + GitHub | Registry + GitHub topics | | awesome-mcp-servers (GitHub list) | Hand-curated, ~1k entries | PR | Your README's first paragraph | | MCPize Marketplace | Curated, growing | Built into MCPize CLI | Your mcpize.json + README | | Smithery (smithery.ai) | ~7k curated | Form + GitHub URL (paid tier for featured) | README + manual edit |

Counts are approximate as of Q2 2026 and grow weekly, treat them as relative ranking, not exact figures.

Strategy: publish to the Registry FIRST (Step 7). Most other directories sync from it, so a single Registry submission propagates downstream within a day or two. Then submit to the four directories that don't auto-sync (Smithery, MCPize Marketplace, awesome-mcp-servers PR, plus a verification check that Glama / PulseMCP picked you up). The full sweep is one focused afternoon.

Schritt 2: awesome-mcp-servers

Two relevant lists (forks fork forks):

  • https://github.com/punkpeye/awesome-mcp-servers (most active)
  • https://github.com/wong2/awesome-mcp-servers (older, large)

Submission is a PR adding one line to the appropriate category. The maintainer's review usually merges within 1-3 days.

- [your-mcp-server](https://github.com/you/your-mcp-server), One-sentence description. Highlights what makes it different from the 5 closest siblings.

Two anti-patterns:

  • Marketing-speak ("revolutionary AI-powered enterprise-grade ..."). Maintainers reject these.
  • Cluttering multiple categories. Pick one. If you don't know which, ask in the PR comments.

The PR description should include: what your server does, who it's for, install command. Three lines max.

Schritt 3: mcp.so submission

https://mcp.so/submit

Form fields (all required):

  • GitHub URL
  • Description (1-2 sentences, mirrors your README first paragraph)
  • Tags (3-5 relevant, e.g. database, claude, automation)
  • Category (pick the closest match)

After submission, mcp.so auto-pulls from your README:

  • Project name + description
  • Install snippet (looks for the first ```bash block with claude mcp add)
  • Tools list (looks for ## Tools heading + table)
  • Demo screenshot (looks for the first ![](url) image in README)

Optimization: if your README's first 200 chars are clear, your mcp.so listing looks great. If it's a wall of badges, it looks confused.

Indexing takes ~24 hours. Re-submission isn't necessary; mcp.so re-fetches your README weekly.

Schritt 4: MCPize Marketplace

If you deployed via MCPize (Phase 8.1), your server is already in the MCPize Marketplace by default. Edit the listing in the MCPize dashboard:

  • Tagline, one sentence, max 80 chars
  • Long description, mirrors README, can be richer (markdown supported)
  • Featured tags, 3-5 keywords
  • Pricing. Free / Paid / Subscription (links to your /upgrade page)

Visibility in MCPize is curated, high-quality servers get featured, low-effort ones rank low. Curators look for: working demo, clean README, recent commits, reasonable response times on /health.

Schritt 5: Smithery

https://smithery.ai

Form-based submission. Reviewer manually approves. Smithery has a paid tier ($X/mo) for "featured" placement, but the free listing is enough to get discovered.

Pulls from README + your repo's GitHub topics. Make sure you've added topics like mcp, claude, model-context-protocol on the GitHub repo (Settings → Topics).

Schritt 6: Glama

https://glama.ai

Auto-indexes any GitHub repo with mcp in the topics + a package.json exposing an MCP-compatible server. No manual submission needed, just make sure your GitHub topics include mcp and model-context-protocol.

Pulls README + tags. Smaller audience, growing fast.

Schritt 7: Official MCP Server Registry

The single highest-credibility placement is the official MCP Server Registry at https://registry.modelcontextprotocol.io/.

Note: the older modelcontextprotocol/servers GitHub repo no longer accepts new third-party server listings, its README explicitly redirects new submissions to the Registry. Don't waste a PR there.

The Registry is a machine-readable index that MCP clients (Claude Desktop, Cursor, Codex) consume programmatically. Listings are authored as a server.json in your repo and submitted via the official mcp-publisher CLI:

# Install the publisher CLI
npm install -g @modelcontextprotocol/publisher  # check the actual package name in the Registry docs

# Auth (one of OIDC via GitHub Actions, npm-package-owner verification, or DNS-TXT)
mcp-publisher login github

# Author server.json next to your package.json (minimal example)
cat > server.json <<EOF
{
  "name": "io.github.you/your-mcp-server",
  "description": "What it does in one sentence.",
  "repository": {
    "type": "git",
    "url": "https://github.com/you/your-mcp-server"
  },
  "version_detail": {
    "version": "1.0.0",
    "release_date": "2026-04-25"
  },
  "packages": [{
    "registry_name": "npm",
    "name": "your-mcp-server",
    "version": "1.0.0"
  }]
}
EOF

# Submit
mcp-publisher publish

After acceptance:

  • Your server appears at https://registry.modelcontextprotocol.io/v0/servers?search=your-server.
  • Downstream marketplaces (mcp.so, MCPize, Smithery, Glama) sync from the Registry, so a single Registry submission propagates everywhere.
  • Updates: bump version_detail.version + re-publish. The Registry treats versions as immutable, published versions don't change retroactively.

Three things the Registry verifies before accepting:

  • The package on npm matches the version in server.json (no drift).
  • The repo URL resolves and points at the same package.
  • Your authentication proves you own either the GitHub repo or the npm package.

This one is worth doing first. The downstream sync means one publish covers most of the marketplace ecosystem in Step 1.

Client-Check · auf Deinem Rechner ausführen
cat package.json 2>/dev/null | python3 -c "import json,sys; p=json.load(sys.stdin); deps=list((p.get(\"dependencies\") or {}).keys()); print(\"sdk:\", \"@modelcontextprotocol/sdk\" in deps); print(\"bin:\", bool(p.get(\"bin\"))); print(\"main:\", bool(p.get(\"main\")))" 2>/dev/null || echo "no package.json in cwd"
Erwartet: sdk: True, plus either bin or main is True.
Falls hängen geblieben: Run `npm init -y && npm install @modelcontextprotocol/sdk zod`, then add `"bin": { "your-server": "./dist/server.js" }` to package.json.

Schritt 8: How to spot directories not worth your time

The MCP space has ~30 directories. Most of them are scrapers that auto-add every npm package matching mcp-*. Indicators a directory isn't worth submitting to:

  • No editorial review, every npm package shows up. Yours will too, no submission needed.
  • No clear submission process (no form, no PR target), they're not actively curating.
  • Domain age < 6 months + no GitHub traffic, speculative listing site, no real audience.
  • "Submit your server, $99 for featured placement", paywalled marketplaces don't have organic traffic.
  • No about / no maintainer name, automated scraper, no human discovery.

Stick to the six in Step 1. Anything else is bonus, not core.

Schritt 9: Verify

Run academy_validate_step. The validator checks package.json is wired.

Manual end-to-end after submissions:

# 1. mcp.so listing exists
curl -s https://mcp.so/server/your-mcp-server | grep -o '<title>[^<]*'
# Should return your server's title

# 2. awesome-mcp-servers PR merged
gh pr list --repo punkpeye/awesome-mcp-servers --state merged --search "your-mcp-server"

# 3. Glama auto-index picked up
curl -s "https://glama.ai/api/mcp-servers?search=your-mcp-server"

# 4. MCPize listing reachable (only if deployed via MCPize)
curl -s https://mcpize.com/marketplace/your-mcp-server

Within a week of submission, all six should be active.

Common traps

  • Submitting to scrapers + paid lists for "exposure", wastes time, attracts no real users.
  • Generic descriptions ("MCP server for X"), mcp.so listings look identical. Differentiate.
  • Forgetting GitHub topics. Glama can't auto-index without mcp topic on the repo.
  • PR to awesome-mcp-servers without the right category, maintainer asks for changes, slows merge.
  • Re-submitting to same directory weekly, annoying to maintainers, doesn't speed indexing.
  • Marketing-speak descriptions, every directory's curators reject these on sight.

What good looks like

Within 7 days of npm publish: mcp.so listed, awesome-mcp-servers PR merged, Glama auto-indexed, MCPize Marketplace edited, Smithery submitted, Anthropic official-list PR open. Discovery traffic starts within 2 weeks. After 60 days, ~30-50% of new installs come via these directories (the rest via direct README share, blog posts, social).

The work is one focused afternoon. The payoff is steady-state organic discovery for as long as your project is maintained.

GitHub OSS, trust signals thatTrusted publishing, token-less