data·vault_
← Catalog

MCP server directory: the registry, five community indexes, and our three servers

Published 2026-09-10. A directory of MCP servers in two layers: where the ecosystem indexes them, and the three servers we publish with the exact config JSON from each README. The first half is deliberately short; the big indexes already do breadth better than we ever will.

Where MCP servers are indexed

An MCP server only exists to a client after it is configured, so discovery happens through indexes. Two kinds matter. The official registry is where servers register under verified names. The community directories crawl it and each other, then add categories, search, and descriptions. If you are hunting for a server for a specific job, check both: the registry for provenance, a directory for browsing.

DirectoryWhat you will find there
MCP Registrythe official registry run by the team behind the Model Context Protocol; servers register under verified names like io.github.jayjex/dataset-mcp, and clients can resolve them programmatically
PulseMCPa large browsable index with categories, per-server tool listings, and weekly coverage of the ecosystem
mcp.soa search-heavy index that lists each server's tools and links straight to the repo
Glamaa directory with per-server tool docs, plus its own MCP client and gateway on the same site
mcpservers.orga curated, categorized listing with community reviews per server

A listing on any of these is a starting point, not a vetting. Before wiring a stranger's server into Claude Desktop or another client, check the license, whether the repo is alive, and what the install command actually runs. Our own MCP tools list shows what a server's tool catalog looks like over raw JSON-RPC, so you can compare before trusting a description.

Our three servers

All three are MIT-licensed, free to run, and work over stdio with no API key. Each entry below uses the config JSON from the server's README, verbatim.

ServerToolsData sourceLicense
dataset-mcp5149,673 rows across 5 datasets, downloaded once from a public GitHub releaseMIT
earn-bounty-scanner3live Superteam Earn API at superteam.fun, fetched per callMIT
solana-research-library425 field-probed platforms, database ships inside the package, no network at runtimeMIT

dataset-mcp

Query the Data Vault catalog from any MCP client: filter and paginate full files with query_dataset(), pull stats with get_stats(), preview with get_sample(). Version 1.1.0, listed in the official registry as io.github.jayjex/dataset-mcp. Full docs on the MCP server for datasets page; every tool with captured output on the MCP tools list.

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "dataset-mcp": {
      "command": "npx",
      "args": ["-y", "github:jayjex/dataset-mcp"]
    }
  }
}

npm 12 refuses git-based installs by default, so on npm 12 use ["--allow-git=all", "-y", "github:jayjex/dataset-mcp"] instead, or set allow-git=github.com in ~/.npmrc. npm 10 and 11 need no flag. TOML-based clients (pi, Codex) use [mcp_servers.dataset-mcp] with the same command and args.

earn-bounty-scanner

Live Solana bounties from Superteam Earn: search with full description text, fetch one bounty by id, or list what is open right now. The agentAccess field tells an agent whether it may submit (AGENT_ALLOWED) or is barred (HUMAN_ONLY). Listed in the official registry as io.github.jayjex/earn-bounty-scanner.

From the README, install and config:

{
  "mcpServers": {
    "earn-bounty-scanner": {
      "command": "npx",
      "args": ["-y", "earn-bounty-scanner"]
    }
  }
}

solana-research-library

A curated intelligence DB of 25 Solana/web3 earning platforms, each field-probed on 2026-09-04 with a verdict (LIVE-RAIL / USER-GATED / CONDITIONAL-PASS / WATCHLIST / DEAD), the exact payout rail, and the KYC gate. The 11 DEAD entries are kill-logged with reasons, so an agent skips re-researching lanes that already failed verification. The database ships inside the package: tools answer instantly and an upstream outage cannot fail a call.

From the README, install and config (npm package pending, so the git form for now):

git clone https://github.com/jayjex/solana-research-library
cd solana-research-library
npm install
npm run build
{
  "mcpServers": {
    "solana-research-library": {
      "command": "node",
      "args": ["/absolute/path/to/solana-research-library/dist/index.js"]
    }
  }
}

Once the npm package is published, npx -y solana-research-library will work too. All three servers in one config block, plus the tool-by-tool outputs: MCP tools list.

Installing them in Claude Desktop

The config JSON above drops into claude_desktop_config.json; after saving, fully quit Claude Desktop and reopen it, because the file is only read at startup. The MCP server for Claude Desktop guide walks the whole path with verified outputs: file locations on macOS and Windows, the npm 12 EALLOWGIT error, and what the tools menu should show once the server connects. A Claude-focused comparison of the same three servers sits on the MCP tools for Claude page.

How this directory is scoped

This page lists servers we build and run, with outputs captured from real sessions, and links out to the indexes that cover the rest of the ecosystem. That is the whole editorial rule: breadth belongs to PulseMCP, mcp.so, Glama, and mcpservers.org; provenance belongs to the official registry; what we can verify belongs here. If a server appears on this page, we have run it over stdio and documented its tools.

FAQ

What is an MCP server directory?

A browsable index of MCP (Model Context Protocol) servers: what each one does, which tools it exposes, and how to install it. The MCP ecosystem has one official registry run by the team behind the protocol, plus several community-built directories that add search, categories, and editorial descriptions on top.

Which MCP server directory is the official one?

The MCP Registry at registry.modelcontextprotocol.io, run by the team behind the Model Context Protocol. It is the canonical place where servers register under verified names like io.github.jayjex/dataset-mcp. The community directories index it and add their own discovery layers, but a listing on any of them is not an endorsement; check the license, the repo activity, and ideally run the server yourself before wiring it into a client.

Are the servers in this directory free?

Yes. All three servers listed here are MIT-licensed, free to run, and need no API key. dataset-mcp downloads its data files once from a public GitHub release, earn-bounty-scanner fetches the public superteam.fun API per call, and solana-research-library ships its database inside the package with no network calls at runtime.