You need to find an MCP server for Slack. Where do you look?

MCQ·Easy·4.0 · 0·~1 min·Asked atC3 Ai Freshworks Midjourney·Relevant atAnthropic

Attempt it

The official MCP Registry (in preview since September 2025) lists server metadata, but it hosts no packages. Four complementary sources cover the rest of the discovery problem, and engineers use whichever fits the moment. **`github.com/modelcontextprotocol/servers`** is the official reference repo maintained by the MCP team. It hosts reference server implementations (Filesystem, Git, Fetch, Memory, Time, and others), and its README points to the MCP Registry for community servers. Start here for trusted, vetted servers. **mcp.so** is a community-curated marketplace organized by category (databases, devtools, communication, AI), with screenshots, install instructions, and reviews. Good for browsing what exists across the ecosystem. **glama.ai/mcp/servers** is a searchable index that tracks usage statistics and version freshness, making it easy to separate actively maintained servers from abandoned weekend projects. **smithery.ai** is an installer-style registry. Pick a server, click install, and it edits your `claude_desktop_config.json` or `.cursor/mcp.json` for you. This is the one-click experience that removes the JSON editing step. None of these hosts the actual server artifacts. The MCP server itself typically lives as an npm package, a PyPI package, or a Docker image. The discovery sites point at those underlying artifacts.

MCP server discovery is intentionally federated. The official MCP Registry (in preview since September 2025) is a central catalog of server metadata, but unlike npm or PyPI it hosts no packages. Four other sources cover the rest of the discovery need. **`github.com/modelcontextprotocol/servers`** is the official reference repo maintained by the MCP team. It hosts reference server implementations (Filesystem, Git, Fetch, Memory, Time, and others), and its README points to the MCP Registry for community servers. This is the trusted starting point and the source most install guides reference. **mcp.so** is a community-curated marketplace. Browse by category (databases, devtools, productivity, AI), see install snippets, screenshots, and reviews. Best for discovering servers you did not know existed. **glama.ai/mcp/servers** is a searchable index emphasizing operational metadata: usage statistics, version freshness, last commit dates. The practical value is filtering for maintained servers in a long tail where many entries have not been updated in months. **smithery.ai** is an installer plus directory. Browse, pick a server, choose your host (Claude Desktop or Cursor), click install, and Smithery edits the right config file on your behalf. Removes the JSON-editing friction for non-technical users. Two structural properties matter. First, these sites aggregate; they do not host artifacts. The actual packages live on npm, PyPI, or Docker registries. Second, they all describe servers built against the same protocol, so a server found in any source works in any compliant host (Claude Desktop, Cursor, Zed, Continue, Cline).

TL;DR

Discovery is federated: the official modelcontextprotocol/servers reference repo plus community indexes like mcp.so and glama.ai and the smithery.ai installer. The official MCP Registry lists metadata, not packages.

Memory aid

Easy to grasp

Imagine you want to find a specific kitchen gadget. There is no single website where every gadget lives. Instead, there is the official toolmaker's catalog (the reference gadgets they made themselves), a couple of independent shopping sites that list many brands and let you read reviews, and a helper that can drop a chosen gadget into your kitchen drawer with one click. MCP server discovery looks exactly like that. The official modelcontextprotocol/servers GitHub repo is the toolmaker's catalog. mcp.so and glama.ai are the shopping sites. smithery.ai is the one-click installer. None of them is the single true registry, and they all describe servers that use the same protocol, so anything you find in one place works in any MCP host.

Key concepts

MCP server registry Smithery installer MCP server discovery

Concept explanation~2 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

'Where do I find an MCP server?' depends on what you mean by find. There is no mcp install foo-server command and no mcp-hub.io package registry. Discovery and installation are federated across several complementary sources, each stronger at a different part of the problem.

The four common sources

Alongside the official MCP Registry, four sources cover discovery. Knowing all four is enough to find any MCP server worth finding.

github.com/modelcontextprotocol/servers is the protocol team's reference repo. It hosts reference server implementations (Filesystem, Git, Fetch, Memory, Time, and others) as worked examples of the SDK. The README points to the MCP Registry for community servers. This is the trusted starting point and the source most install guides link to first.

mcp.so is a community-curated marketplace. Browse by category (databases, devtools, communication, productivity, AI). Each entry shows screenshots, install instructions, README excerpts, and reviews. Best for discovering 'what kinds of servers exist that I might not have considered.'

glama.ai/mcp/servers is a searchable index emphasizing operational metadata: usage statistics, version freshness, last commit dates. The value is filtering for actively maintained servers in a long tail where many entries are abandoned side projects.

smithery.ai is the install-layer aggregator. It overlaps with the marketplaces for browsing, but its differentiator is one-click installation into a target host. Pick a server, click install, choose Claude Desktop or Cursor, and Smithery edits the right config file on your behalf. This removes JSON-editing friction for non-technical users.

None of these is canonical. They overlap. Most experienced engineers check one or two depending on what they need, defaulting to the official repo first.

Aggregators, not artifact hosts

Why discovery is federated, not centralized

Cross-host portability as the unifying property

Where discovery is heading

Situations where this technique stops working.

2–4 min · Everything important, quickly.

Real products, models, and research that use this idea.

An engineer looking for a Git MCP server checks github.com/modelcontextprotocol/servers first, finds the official reference implementation, and copies the config snippet into claude_desktop_config.json.
A non-technical user installs the Linear MCP server via smithery.ai with one click. Smithery edits the Cursor config file behind the scenes and the user never touches JSON.

What an interviewer would ask next. Try answering before peeking at the approach.

QWhy has the MCP ecosystem not converged on a single package registry like npm?

The ecosystem is young (launched November 2024). The install model is JSON-edit-and-restart, not 'npm install', so there is no unifying install command to centralize around. Fragmentation lets multiple registries experiment with curation, ratings, and install flows in parallel. The official MCP Registry (preview since September 2025) centralizes metadata, while packages stay on npm, PyPI, and Docker.

2 more follow-ups an interviewer would ask next. Sign in to reveal them.

Red flags & common mistakes

The phrases that signal junior thinking. Click to expand.

Most common mistake

Expecting one canonical registry like npm that hosts MCP server packages. Discovery is intentionally federated across several sources; the actual artifacts live on npm, PyPI, or Docker, not on the discovery sites.

60 second bullets to scan on the way to the call.

What the official MCP Registry provides and what it does not host
The four common discovery sources and what each is best at

Primary sources. Browse if you want the original framing.