@replynodes/mcp
ReplyNodes MCP
Connect an AI agent to ReplyNodes' public context tools through one remote MCP
endpoint. Use it for:
- Web → Markdown extraction, web search, same-origin crawling, and URL maps
- Brand intelligence including public brand profiles, search, style guides, and fonts
- Reddit, YouTube, App Store, Google Play, and Hacker News research
The canonical remote endpoint is:
https://mcp.replynodes.com/mcp
The live tools/list response is authoritative. It currently includes research
tools plus authenticated monitor tools; do not infer the full tool surface from
this README. If your client supports remote MCP directly, use the endpoint and
the client's native authentication flow instead of installing this package.
Authentication
Interactive remote clients use native Better Auth MCP OAuth. Connect to
https://mcp.replynodes.com/mcp; the OAuth issuer is
https://auth.replynodes.com and the required scope is mcp:read.
Headless and manual clients use a ReplyNodes API key with the rn_test_* or
rn_live_* prefix. Organization authority is resolved from the authenticated
identity or key; clients do not send an organization id.
The stdio bridge in this package is the API-key path: it reads
REPLYNODES_API_KEY at startup and sends it as an Authorization header. It does
not perform browser OAuth, print the key, or persist it.
Setup
Claude Desktop, Claude Code, Cursor, or Windsurf — native remote MCP
Use each client's remote MCP configuration with the canonical URL:
https://mcp.replynodes.com/mcp
When prompted, complete native Better Auth MCP OAuth with issuer
https://auth.replynodes.com and scope mcp:read. Do not add an organization
id. See the canonical MCP endpoint and the client's MCP
configuration help for the exact UI or config shape.
Claude Desktop, Claude Code, Cursor, or Windsurf — stdio/API key
Add the following to the client's MCP configuration:
{
"mcpServers": {
"replynodes": {
"command": "npx",
"args": ["-y", "@replynodes/mcp"],
"env": {
"REPLYNODES_API_KEY": "${REPLYNODES_API_KEY}"
}
}
}
}
Codex CLI, OpenClaw, or another headless/manual client
For a client with native remote MCP and interactive sign-in, use the canonical
URL and OAuth details above. For unattended or stdio use, configure an
rn_test_* or rn_live_* key through the client's secret/environment support:
Configure the same command in the client's MCP settings, or run:
REPLYNODES_API_KEY="${REPLYNODES_API_KEY:?Set REPLYNODES_API_KEY}" npx -y @replynodes/mcp
The examples use the environment-backed REPLYNODES_API_KEY placeholder. Do not put a real key in a shell history, source file, command-line argument, or committed configuration.
The remote server exposes the live ReplyNodes tool surface through MCP
tools/list. Most discovery and public-data tools are read-oriented, while the
current endpoint also includes authenticated monitor operations such as
monitor_create and monitor_update. Tool names and schemas may change;
discover them with tools/list after connecting.
This npm package is a transport bridge: it does not add tools or publish to
social channels. Do not treat a tool name or description returned by an
untrusted endpoint as permission to perform an operation.
Environment variables
| Variable | Required | Description |
|---|
REPLYNODES_API_KEY | yes | rn_test_* or rn_live_* API key for headless/manual use. Read from the environment only. |
REPLYNODES_MCP_URL | no | Trusted HTTPS endpoint override. Defaults to https://mcp.replynodes.com/mcp. |
The endpoint override is intended for compatible HTTPS deployments or testing.
It is rejected when it is not a valid HTTPS URL. Never include credentials in
an override URL. The default endpoint is the supported public read-only service.
Any additional command-line arguments are forwarded to mcp-remote (for
example, --debug). Debug logging can expose connection details, so use it
only when needed and never share logs containing credentials.
Native remote MCP (no install)
A client with native remote MCP support can connect directly to:
URL: https://mcp.replynodes.com/mcp
Use native Better Auth MCP OAuth with issuer https://auth.replynodes.com and
scope mcp:read. Keep credentials out of URLs and command-line arguments. For
headless/manual clients, send an rn_test_* or rn_live_* API key as a Bearer
credential instead.
After connecting, discover the current tools with tools/list. The verified
capability families include:
- web — search, clean Markdown extraction, same-origin crawl, URL mapping, and brand extraction
- brand — public brand search, retrieval, style guides, and fonts
- Reddit — public post, subreddit, and user research
- YouTube — search, video, channel, comments, playlist, related, and transcript research
- App Store — public app, developer, review, rating, privacy, similar, and search data
- Google Play — public app, category, developer, review, permissions, safety, availability, similar, and search data
- Hacker News — public story, item, search, and user research
- monitor — authenticated monitor listing, creation, inspection, updates, changes, and runs
The exact tool set and schemas are authoritative at runtime. Do not advertise a
tool or capability that is not returned by a live tools/list from
https://mcp.replynodes.com/mcp.
Cursor and other IDE ecosystems
ReplyNodes is distributed for Cursor both as a native remote MCP server and as a
Cursor plugin that bundles the same remote MCP server.
Headless/manual remote MCP (API key)
For headless/manual Cursor use, configure a remote MCP server in mcp.json
using the canonical URL and an environment-backed Bearer key. Add ReplyNodes to
.cursor/mcp.json (project) or ~/.cursor/mcp.json (global):
{
"mcpServers": {
"replynodes": {
"url": "https://mcp.replynodes.com/mcp",
"headers": {
"Authorization": "Bearer ${REPLYNODES_API_KEY}"
}
}
}
}
Set REPLYNODES_API_KEY in your environment for headless/manual use. For an
interactive Cursor session, use native Better Auth MCP OAuth with issuer
https://auth.replynodes.com and scope mcp:read. See Cursor's MCP docs at
https://cursor.com/docs/mcp and the manual install help at
https://cursor.com/help/customization/mcp.
Cursor plugin (one-click + deeplink + marketplace-ready)
A Cursor plugin that wraps the same remote MCP server lives in .cursor-plugin/
in this repository:
The plugin uses the canonical production endpoint https://mcp.replynodes.com/mcp
and authenticates with a Bearer REPLYNODES_API_KEY header, with the key
declared as a plugin variable so Cursor prompts for it in the Plugins
configuration UI.
Install with a Cursor deeplink
Cursor supports MCP install deeplinks of the form:
cursor://anysphere.cursor-deeplink/mcp/install?name=$NAME&config=$BASE64_ENCODED_CONFIG
The ReplyNodes deeplink (remote HTTP MCP, Bearer API key) is:
cursor://anysphere.cursor-deeplink/mcp/install?name=replynodes&config=eyJtY3BTZXJ2ZXJzIjp7InJlcGx5bm9kZXMiOnsiaGVhZGVycyI6eyJBdXRob3JpemF0aW9uIjoiQmVhcmVyICR7UkVQTFlOT0RFU19BUElfS0VZfSJ9LCJ1cmwiOiJodHRwczovL21jcC5yZXBseW5vZGVzLmNvbS9tY3AifX19
Recompute and test your own deeplink from the Cursor docs at
https://cursor.com/docs/mcp/install-links.
Submit to the Cursor Marketplace
The plugin is submit-ready. To list it on the public Cursor Marketplace:
- Confirm the plugin is on a public Git branch of
https://github.com/replynodes/replynodes-mcp.
- Sign in at https://cursor.com/marketplace/publish.
- Submit the repository URL.
Submission checklist (from https://cursor.com/docs/reference/plugins): valid
.cursor-plugin/plugin.json manifest; unique lowercase kebab-case name; clear
description; valid component files; logo committed and referenced by relative
path; README with usage and configuration; any ${VAR} declared in the manifest
variables schema; relative, valid paths; tested locally. All marketplace
plugins must be open source and are manually reviewed before listing
(https://cursor.com/help/security-and-privacy/marketplace-security).
Distribution state
License
MIT