20 newer servers that their makers sent us. We have read what each one says it does and written it up in plain words, but we have not installed them yet, so they sit here rather than on the main shelf.
Before you add one
Hosted servers see whatever you send them. Check who runs it and what the sign-in can reach.
Servers that run on your computer can do what your user account can. Read the repository first.
When a server passes our own install check, it moves to the main shelf with a full page.
Looks up the official docs for the exact library version your code uses.
Try asking
“How do I validate search params in TanStack Router? Quote the docs for the version in my package.json.”
A coding agent asks Docsy which library and version it is dealing with, then pulls the matching passages from the official documentation, each with a link back to its source. All five of its tools only read.
Endpoint:
https://docsy.remcp.site/mcp/registry
Before it works:
Sign-in through OAuth the first time a client connects.
“Start a GPU, run train.py from my workspace, and stop the machine when it finishes.”
The assistant can start a GPU machine in your Hinode workspace, run a job, read the log, copy files either way and pause the machine afterwards. It reads the current price before starting anything you pay for and can tell you what a run cost.
Before it works:
A Hinode account, signed in through OAuth. GPU time is billed.
GitHub Trending by day, week or month, with a short review of each project.
Try asking
“What Rust projects are climbing GitHub Trending this week, and why do people like them?”
Ask what is climbing GitHub Trending this week, in one language or on one topic, and HotRepo answers with the repositories, their numbers and a written review. It can also search past boards and its monthly round-ups.
Endpoint:
https://hotrepo.ai/mcp
Before it works:
Needs an API key, made after signing in with GitHub, sent as a Bearer header. The free plan caps calls per day and keeps a week of history.
A second opinion an agent can ask for before a risky step.
Try asking
“Before you delete these 40 files, ask Jev whether that matches what I asked for.”
Before an agent deletes something, sends an email or calls a task finished, it can put the question to Jev Agent and get back a yes or no, a pick between options or a score, each with a probability. The agent can then stop or try again.
Before it works:
A handful of free credits each month, paid after that.
Lets a coding assistant run shell commands on a different computer.
Try asking
“On my build box, check free disk space and tail the last 50 lines of the deploy log.”
The server runs next to your assistant and passes commands through an HTTP relay to a second machine, which listens with a curl-and-bash script, so there is no SSH or agent to install there. Its three tools run a command, show the session and reset it.
Start command:
npx -y remotify-mcp@latest
Before it works:
Free and open source (Apache-2.0). Needs Node.js 18 or later; the relay can be self-hosted.
Worth knowing:
This gives the assistant a shell on another machine. Keep the mode where you approve each command.
Quick sanity checks on a phone number or an email address.
Try asking
“Are these five signup emails from throwaway domains?”
Two tools. check_phone reads the numbering plan to say whether a number is well formed, which country it belongs to and what kind of line it is. check_email looks at the syntax and the domain: mail records, throwaway providers and role addresses such as info@.
Before it works:
Free beta by invitation, with a fixed number of checks per key.
Worth knowing:
It does not ask the carrier or the mail server, so it cannot tell you a number or inbox is live.
Lets ChatGPT or Claude use the Chrome you are already signed in to.
Try asking
“In my Chrome, open the supplier portal and fill in the reorder form from this list.”
You install the BrowserPair extension and pick which sites the assistant may touch. It can then open pages, read them and fill in forms in your own browser, and it stops to ask before anything sensitive. Each task leaves a record of what happened.
Before it works:
The Chrome extension plus an OAuth sign-in from the assistant.
Search a library of AI how-tos drawn from popular GitHub projects.
Try asking
“Show me how people use AI agents to triage GitHub issues, with a prompt I can copy.”
Each entry is a task someone solved with AI, broken into steps, with a prompt you can hand to an agent. The server also returns AI repositories that are trending and a daily news digest.
“What has been in the news in Nairobi this month, grouped by storyline?”
Mappit sorts local news by where it happened. An assistant can pull the stories for a city, follow a running storyline, search the whole archive, or ask questions of its datasets on economies, health, conflict and elections.
Search scholarly papers, authors and institutions in OpenAlex.
Try asking
“Find the most-cited papers on retrieval-augmented generation since 2023 and who cites them.”
Seven read-only tools find papers with their abstracts, walk the citations and references of a paper, and look up authors and institutions in the open OpenAlex index. Answers come back as short JSON.
Turns any public web page into Markdown, or Markdown plus structured fields.
Try asking
“Pull the price, title and stock status from this product page as JSON.”
page_markdown hands back the readable text of a URL. extract_url adds JSON in a fixed shape: article details, a product and its price, or the tables on the page.
Endpoint:
https://api.webextract.dev/mcp
Before it works:
No key to connect. A small number of free calls per day for each IP address, then paid credits; a fetch that fails is not charged.
Scraping, search, site maps and full crawls behind one hosted service.
Try asking
“Map every URL under docs.example.com and fetch the ten newest pages.”
XCrawl is a paid scraping platform, and its MCP server gives an assistant the same four jobs as its API: fetch a page, search, list the URLs of a site, or crawl it. It also ships a CLI and n8n nodes.
Amazon product, keyword and market numbers for research.
Try asking
“How crowded is the Amazon US market for silicone baking mats, and what do the top sellers earn?”
An assistant can look up Amazon products and keywords, size up a market, and read reviews, traffic and estimated sales through the same operations as the REST API.
Before it works:
A paid API.
Worth knowing:
Sales figures are estimates, and not every operation covers every marketplace.
“Which of my pages lost the most clicks in the last 28 days, and for which queries?”
Connect the Search Console properties you choose and ask about clicks, impressions, positions and click-through rate, or what it has seen of indexing and sitemaps. It also does keyword, competitor and backlink research.
Reference answers about ISO 20022 payment messages and codes.
Try asking
“Which ISO 20022 message replaces MT103, and which schemes use it?”
Ask which ISO 20022 message takes over from an old one such as SWIFT MT103, which payment schemes use a message, or what a Bank Transaction Code means. Every answer says which data release it came from, which matters for payments and treasury work.
Ask about your net worth and update balances by chat.
Try asking
“How much has my net worth changed since January, in euros?”
NetworthTrackr is a tracker you fill in by hand, in any of about thirty currencies. Through MCP the assistant can answer questions like how much your net worth moved since January, converted to the currency you ask for, and it can record balances or add a holding when you tell it to.
Before it works:
A free account, approved once from the assistant; it gets a token it can lose, never your password.
Worth knowing:
There is no bank connection. Deleting an account needs its exact name.
Long-term memory for an agent, kept encrypted on your own machine.
Try asking
“What did I decide about the pricing page in last week’s sessions?”
Vaultic keeps conversation transcripts, memories you can search by meaning or by keyword, and a graph of people and how they relate, all in one encrypted local file. Corrections are recorded rather than written over, and there is an audit log.
Before it works:
Nothing leaves your computer; there is no cloud part.
Worth knowing:
It exposes around a hundred tools. Clients that list every tool up front may feel slower with it.
Perfume notes, families and similar scents, from an editor-built index.
Try asking
“I like Terre d’Hermès. What should I try next that is a bit fresher?”
Ask what is in a well-known perfume, what it smells close to, or what to try instead of it. Answers come from an index written by 10ML’s editors and point to their guides.
Furnished flats and rooms in Berlin, from several sites at once.
Try asking
“Find a furnished one-bedroom in Berlin under €1,500 from March, near Alexanderplatz.”
For anyone moving to Berlin: give a budget, a move-in date, how long you are staying and where you need to be, and the assistant searches several rental sites together and compares what comes back. You apply on the original site.
