MCPExplorer
Search, vet & assemble MCP servers from your agent: verified tools, risk labels, and trust scores.
Open source Repository Open in the app JSON README (API)
About
Search, vet & assemble MCP servers from your agent: verified tools, risk labels, and trust scores.
Details
- Kind
- MCP servers
- Topic
- No topic detected
- Publisher
- mcpexplorer
- Origin
- official
- Category
- ferramentas
- Transport
- http
- Version
- 1.0.0
- Stars
- 1
- Last push
- 2026-07-07T12:01:47Z
- Repository state
- ativo
- License
- MIT
- Added
- 2026-08-29 04:00:46
- Updated
- 2026-08-29 04:00:46
- Origin id
io.github.mcpexplorer/mcpexplorer
README
# MCPExplorer
**The search engine and trust layer for MCP servers.**
### → [mcpexplorer.com](https://mcpexplorer.com)
[](https://mcpexplorer.com)
[](https://mcpexplorer.com/mcp)
MCPExplorer indexes [Model Context Protocol](https://mcpexplorer.com/what-is-an-mcp-server)
servers and **verifies each one with a live handshake** — so you can see the real
tools a server exposes, each labeled by risk, plus a trust score, *before* you
install it.
> Every "awesome MCP" list is a static, unverified README. MCPExplorer actually
> runs the servers.
This repo is the public home for MCPExplorer's docs and its MCP server manifest.
The product itself lives at **[mcpexplorer.com](https://mcpexplorer.com)**.
---
## The problem
As the MCP ecosystem grew past thousands of servers, the lists stopped being
useful. From a name and a one-line description you can't tell:
- what tools a server *actually* exposes,
- whether those tools are read-only or can delete your data,
- whether it needs credentials,
- or whether it's still maintained —
until you install it and hope.
## What MCPExplorer does
- **Verifies, doesn't just list.** It runs the real MCP `tools/list` handshake on
every server — remote servers over the network, local/stdio servers in a
hardened, ephemeral Docker sandbox — and extracts the true tool surface.
- **Labels tool risk.** Every tool is tagged **read / write / destructive**, so you
know what a server can do before you connect it.
- **Scores trust.** A deterministic score from provenance, verification, adoption,
and freshness — with every contributing factor shown on the page, not a black box.
- **Stays fresh.** Every fact carries provenance and a "last checked" timestamp.
**Currently ~1,000 servers indexed · 3,500+ tools extracted · 278 handshake-verified.**
## Start here
- 🔍 **Explore the index** — <https://mcpexplorer.com/explore>
- ⭐ **Best MCP servers** (ranked) — <https://mcpexplorer.com/best-mcp-servers>
- 📖 **What is an MCP server?** — <https://mcpexplorer.com/what-is-an-mcp-server>
- 📊 **State of MCP report** — <https://mcpexplorer.com/reports>
---
## The MCPExplorer MCP server
MCPExplorer exposes its own MCP server, so you can **search, vet, and assemble
servers from inside your agent** — without leaving your client. Hand `plan_toolset`
a plain-language goal and get back a recommended, governed toolset.
**Endpoint:** `https://mcpexplorer.com/mcp` (Streamable HTTP, no auth)
**Add it to Claude Code:**
```bash
claude mcp add --transport http mcpexplorer https://mcpexplorer.com/mcp
```
**Or add it to any MCP client:**
```json
{
"mcpServers": {
"mcpexplorer": {
"type": "streamable-http",
"url": "https://mcpexplorer.com/mcp"
}
}
}
```
**Tools (14):**
| Tool | What it does |
|---|---|
| `search_servers` | Search the index by capability, transport, and risk |
| `get_server` | A server's verified tools, install command, and trust breakdown |
| `get_capability` | Servers that provide a given capability |
| `compare_servers` | Side-by-side comparison of two servers |
| `get_alternatives` | Servers similar to a given one |
| `check_trust` | Explainable trust score and reasoning for a server |
| `generate_runtime_config` | Ready-to-paste client config for a server |
| `list_capabilities` | The capability slugs you can search and look up by |
| `list_runtimes` | Runtimes `generate_runtime_config` supports |
| `get_ecosystem_stats` | Live totals + provenance/transport breakdowns |
| `list_loadouts` | Curated server kits for a job (GTM, coding, research, …) |
| `get_loadout` | One kit with live trust + aggregated tool-risk |
| `plan_toolset` | Plain-language goal → a recommended governed toolset |
| `report_broken_server` | Flag a stale or broken listing |
---
## How verification works
Every server flows through a crawler pipeline:
```
discover → fetch → parse → normalize → resolve → enrich →
classify → verify (live handshake) → score trust → decide indexability
```
The **verify** stage is the heart of it. Remote servers are probed over the
network; local/stdio servers are executed in a locked-down container
(`--read-only`, `--cap-drop ALL`, `--security-opt no-new-privileges`, no host
mounts, no secrets, memory/CPU/PID limits, auto-removed) that runs only known
launchers (`npx`/`uvx`/`uv`/`pnpm`/`bunx`). The handshake extracts the real
`tools/list`, and each tool's risk is derived from its name and schema.
A page is only indexed once it has real MCP-server evidence **and** enough
verified facts — so the index stays servers, not framework repos or link lists.
## For agents & programmatic access
- **MCP server:** `https://mcpexplorer.com/mcp` (see above)
- **llms.txt:** <https://mcpexplorer.com/llms.txt>
## Submit a server
Missing something? Submit it at <https://mcpexplorer.com/submit> — it enters the
same verification pipeline as everything else.
## FAQ
**Is this affiliated with Anthropic?**
No. MCP is Anthropic's open protocol; MCPExplorer is an independent index built on
top of it.
**Is the trust score a black box?**
No. It's deterministic, and every `+`/`−` factor is shown on each server's page.
**How do you run untrusted servers safely?**
Local/stdio servers only ever run inside the hardened, ephemeral container
described above — never on the host, never with secrets.
---
<sub>MCPExplorer · [mcpexplorer.com](https://mcpexplorer.com) · a discovery, trust,
and capability graph for MCP servers.</sub>