find-mcp-directories

Drive the ServiceGraph API (https://api.servicegraph.co) to find and rank directories and registries where a builder can list an MCP server (Model Context Protocol) via the product_directory dataset. The catalog has 1,000+ directories; ~20 are dedicated MCP registries (e.g. registry.modelcontextprotocol.io, smithery.ai, mcp.so, glama.ai, pulsemcp.com), each enriched with Domain Rating (dr), backlinks, and organic traffic.

This is a "where to publish / where to get backlinks" skill, not a "who to hire" skill. Each row is a registry you submit to, not a firm and not a server. The payoff is a backlink from a high-authority domain (SEO) plus discovery by agent builders browsing for servers.

Any HTTP client works (curl, fetch, requests). Examples below use curl.

Sibling skills — defer when the niche is broader

• General AI tools / AI agents / agent skills ("where to list my AI tool", "agent-skills directories") → find-ai-directories. (Several registries list MCP servers and skills/tools — if the artifact is specifically an MCP server, this skill is the right pick.)
• General SaaS / software / app launch ("Product Hunt alternatives") → find-product-directories.

MCP server (preferred for authed calls)

If your harness has the ServiceGraph MCP server loaded (tool names contain servicegraph), prefer those tools — credentials stay in the harness sandbox via OAuth 2.1 + PKCE, no token in LLM context. Otherwise use the REST flow.

API surface (dataset id: product_directory)

Every endpoint requires the bearer (Authorization: Bearer vk_…).

Cost model. Discovery / search / brief reads are free — including the dr ranking signal. Unlocking a row costs 10 credits, lasts 30 days, and reveals the gated fields: editor_note (how to submit + whether the listing gives a backlink), organic_traffic, and total_visits.

Auth

Tokens are vk_* API keys. Keep the token out of the LLM context — never read .env* into context; route authed calls through a shell wrapper.

1. Try the call through a wrapper that sources .env.local:

   ( set -a; [ -f .env.local ] && . ./.env.local; set +a;
     curl -sS -H "Authorization: Bearer $SERVICEGRAPH_API_KEY" \
          'https://api.servicegraph.co/v1/datasets/product_directory/fields' )
   

2. On 401, prompt the user (don't accept the key in chat):

   "Open https://servicegraph.co/profile/api-keys, sign in, create a key, and add SERVICEGRAPH_API_KEY=vk_… to .env.local (or export it). Tell me when done — please don't paste the key into chat."

3. Retry after the user signals ready.

Filter DSL

GitHub-search-style. AND binds tighter than OR; comma list = OR within one predicate; negation is -x / NOT x; any bareword is a keyword search across the directory's name, title, description, listed metadata, and the niche tag (so mcp matches dirs whose niche is "MCP servers"). Multiple barewords AND; wrap multi-word phrases in double quotes.

mcp dr>=60
mcp OR "model context protocol"
mcp servers          # → keyword:mcp AND keyword:servers

Fields that matter here

Because dr is free and briefs are pre-sorted by it, rank a shortlist by authority for zero credits — unlock only to reveal submission notes + traffic.

Identifying rows — apex

Keyed by apex domain (smithery.ai, not a full URL; subdomains like registry.modelcontextprotocol.io are kept as-is when that's the catalog key). Strip user-supplied URLs before :apex or unlock calls.

Recipes

A. The MCP registry shortlist

User: "Where should I list our new MCP server?"

GET /v1/datasets/product_directory/search?filter=mcp&limit=20
# → registry.modelcontextprotocol.io (dr 90), smithery.ai (75), glama.ai (72),
#   mcp.so (72), cursor.directory (69), pulsemcp.com (68), … sorted by dr desc

Present the top N by dr (free).

B. High-authority only (backlink quality)

User: "Only the registries with real domain authority for SEO."

GET /v1/datasets/product_directory/search?filter=mcp+dr>=60&limit=15
# → the ~7 strongest MCP registries by DR

C. Rank by real reach (organic traffic)

organic_traffic is gated but filterable — gate to surface high-traffic registries, then unlock to see the numbers:

GET /v1/datasets/product_directory/search?filter=mcp+organic_traffic>=2000&limit=15

D. Broaden — MCP + agent tooling

Some registries list MCP servers alongside agent skills/tools. Cast wider when the strict-MCP pool is thin:

GET /v1/datasets/product_directory/search?filter=mcp OR "model context protocol" OR (agent skills)&limit=20

If the user's artifact is broader than MCP (general AI tool / agent skill), defer to find-ai-directories.

E. Unlock submission instructions for the picks

# Present briefs ranked by dr (free). "Unlocking 6 = 60 credits, 30-day TTL —
# reveals each one's submission note (how + backlink yes/no) and traffic."
POST /v1/datasets/product_directory/unlocks
  { "apexes": ["smithery.ai", "mcp.so", "glama.ai", "pulsemcp.com", "cursor.directory", "registry.modelcontextprotocol.io"] }

Surface each editor_note verbatim — MCP registries vary a lot in submission mechanics (some take a CLI/PR, some a web form) and in whether the listing actually grants a backlink. The note tells the user before they spend effort.

F. BYO apex list — score MCP registries I already have

1. GET /v1/datasets/product_directory/:apex per domain — free brief with dr (404 = not in catalog, no charge). Flag misses, rank hits by dr.
2. User picks N. POST /unlocks (10×N credits, atomic) reveals notes + traffic.
3. Within 30-day TTL, repeat unlocks are free.

Gotchas

• Rows are registries, not firms or servers. Building an MCP server → find-software-developer / find-ai-consultancy. "How does MCP work" is a DIY question, not this dataset.
• dr is free and briefs are pre-sorted by it — rank for zero credits.
• Gated fields are still filterable (organic_traffic>=1000).
• Catalog is global, not US-only. Don't refuse non-US asks.
• The strict-MCP pool is small (~20). If the user wants more reach, broaden to agent-skill/AI-tool registries (Recipe D) or defer to find-ai-directories.
• Multi-word phrases must be quoted. model context protocol = three AND'd keywords; "model context protocol" is one phrase.
• Unlock is atomic (402 charges nothing) and within-TTL re-views are free.

Errors

JSON envelope: {"error": {"code": "...", "message": "..."}}.

End-to-end example

User: "We just shipped an MCP server. Find the highest-authority registries to list it on, and tell me how to submit to the top five."

# 1. Discover (once per session)
GET /v1/datasets/product_directory/fields?include_values=1

# 2. Validate + scope (free)
GET /v1/datasets/product_directory/check?filter=mcp+dr>=50

# 3. Search briefs (free) — pre-sorted by dr desc
GET /v1/datasets/product_directory/search?filter=mcp+dr>=50&limit=20

# 4. User picks 5. "Unlocking 5 = 50 credits, 30-day TTL — reveals submission
#    notes (how + backlink yes/no) and traffic."

# 5. Atomic unlock (charges 50 credits)
POST /v1/datasets/product_directory/unlocks
  { "apexes": ["smithery.ai", "mcp.so", "glama.ai", "pulsemcp.com", "cursor.directory"] }

# 6. Surface each editor_note verbatim + traffic so the user can prioritize.