Bring the part that moves into the chat you already use.
We kept seeing the same setup every time we talked with a CEO or a small business owner. They are using Claude as the model of choice with an Obsidian markdown knowledge base underneath it. That setup works great, until you want to execute on your strategy. We identified two key problems. This is what we built to fix them.
First, what is an MCP server?
MCP stands for Model Context Protocol. It is an open standard for giving an AI model a way into a system it does not already know about, and Claude supports it in both the desktop app and Claude Code.
On its own, Claude knows what is in your conversation and whatever you have pasted into it. Connect an MCP server and Claude can go and fetch what it needs while you are talking to it. You never call a tool yourself and you never see the plumbing. You just ask, and the answer arrives built on data that was pulled a second ago.
You ask, in plain English
"What is rising in search demand this month, and what should I do about it?" No commands, no syntax.
Claude works out what it needs
It picks from the tools we publish, calls them, and can call several before it answers you.
We hand back real rows
Records with their source, date and link attached. Claude reads those and answers you from them.
The right approach, the wrong data structure
Claude and Obsidian is a perfect combination for a wiki knowledge base. We are not replacing that, we are powering it.
A wiki doesn’t understand time.
An internal wiki (Obsidian) is not a time series database and was never trying to be one. It tells you what was true the day somebody wrote the page. It knows the conclusion of data analysis, but not the raw ingredients. So ask it what is rising in search demand this month, or which of your terms lost ground since April, and there is nothing in there to answer with. Benchmates is pulling your data sources in over time, which makes it a multi layered time series. That is the part a static wiki knowledge base cannot do.
Personalized, powered data.
Multiple people inside a business each end up with their own prompts, workflow, knowledge vault to help them do their job. Every AI knowledge base tailored to them is genuinely useful to the person who uses it. But how do you inject that with real-time information to be able to make decisions and act?
Knowledge base plus time series research is a perfect union.
You keep the workflow you already have. Claude stays Claude, Obsidian stays Obsidian. The Benchmates MCP server brings live data sources into that same conversation, so when you are talking to Claude you are grounded with your accurate up to the minute research data. Your wiki brings the branding and the internal strategy. We bring the part that moves. We collect your data every day, and you just have to do what you are already doing. Chat, but now with insight.
What it can pull in.
Fifteen tools. Ask something broad first, then go down into the evidence underneath it.
What people are searching
The SERP data: what your customers are typing at any given point in time, per term, with real week on week and 60 day movement and the queries rising underneath it.
What people are saying
Your Reddit feeds, X, YouTube, news, and the reviews on Yelp and TripAdvisor. Scored into an index per topic, and you can drop straight down into the actual posts with author, date and link.
What is moving on social
Your TikTok and Instagram feeds, snapshotted every day and ranked by how fast a post is moving rather than how big it already got. Something that has doubled since this morning beats something that peaked last week.
Your research vault
Search across every research document we hold for you, and pull any of them up in full. Search hands back documents to read rather than an answer it made up from previews.
The analysis we run on it
The verdict on each topic, the recommendations with the evidence attached, and the content we are proposing off the back of them.
Actions you can take
From the Claude app or the dashboard or your daily email, in a single chat. Open with this and then drill into whatever looks interesting.
It hands back records, not conclusions.
Every mention, post and action comes back with its source, its date and its url on it, so anything Claude tells you can point at the thing it came from. Document search hands over documents and tells Claude to go read them. Your Claude does the thinking. We are the source of truth it thinks with.
One more thing I care about. When there is nothing to show you, the tools tell you which kind of nothing it is. “We have not run that collection yet” and “we ran it and nobody mentioned you” look exactly the same on a chart and mean opposite things. Quiet that looks like calm is the most expensive thing that can go wrong here, so every one of them says which it is.
It only reads
All fifteen tools read. Nothing that comes through here can change anything in your project.
A key belongs to a person
Not to a project and not to a team. Revoke one key on its own without touching anybody else, and which projects it can reach gets worked out on every single call. Take someone off a project and their next question is already scoped right, with nothing to reissue.
We cannot see your session
MCP hands us the name of a tool and its arguments. That is all it hands us. Your conversation, your prompts and whatever you work out never reach our backend.
The numbers cannot drift
These tools read the same queries your digest email reads. Nothing here recalculates a number the pipeline already worked out, so what Claude tells you and what your inbox tells you agree.
Connecting takes about five minutes.
Benchmates issues you an API key. You add one connector in Claude Desktop, or run one command in Claude Code, and then you ask it something. The full guide comes with the key and covers both.
claude mcp add --transport http benchmates \
https://your-server-address/mcp \
--header "Authorization: Bearer YOUR-KEY"Claude Desktop takes the same thing through Settings, then Connectors, then Add custom connector.