MCP server

Ask your own AI assistant about your business

Connect Claude, Cursor, VS Code or any assistant that supports MCP. It can search your records, report data and chats in analysis, calculate totals and link each answer back to the message, file or ERP record it came from. Access is read-only and limited to one workspace.

Connect in two minutes

1
Create a token
In DataLense, open Settings → AI assistants (MCP) and choose New token. Name it after where you'll use it, pick when it expires (30 days, 90 days or 1 year) and copy it. It's shown once.
2
Add DataLense to your assistant
Paste the server address and the token into your assistant's MCP settings. The examples here are ready to copy.
3
Ask
Ask the way you'd ask a colleague. The assistant answers with links back to the record, message or file in DataLense.

Server address

https://my.datalense.app/api/mcp

Streamable HTTP. Send the token as a Bearer token in the Authorization header.

Claude Code

Run in a terminal:

claude mcp add --transport http datalense https://my.datalense.app/api/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

Cursor

Add to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "datalense": {
      "url": "https://my.datalense.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

VS Code with Copilot

Add to .vscode/mcp.json, then start the server from the file:

{
  "servers": {
    "datalense": {
      "type": "http",
      "url": "https://my.datalense.app/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Windsurf, JetBrains and other clients take the same address and header; see their MCP settings. Replace YOUR_TOKEN with the token you copied.

What the assistant can do

Seven read-only tools. The assistant picks them on its own; you just ask in plain words.

list_datasets
The report data read from files shared in chats: each dataset's fields, how many records it has and the dates it covers.
“What report data do we have from the dispatch group?”
search_records
Finds records by words, names, codes or numbers, across ERP documents and rows read from shared reports.
“Find the sales order for Shree Traders from last week.”
get_record
One record in full: every field, how it changed over time, the chats and files it was seen in, and linked records.
“How has the rate on order SO-1182 changed?”
aggregate_records
Totals, counts, averages and trends, calculated in the database rather than by the assistant. Filter and group by any field, or by day, week or month.
“Total dispatch quantity per customer this month.”
search_messages
Finds messages by words in chats set to Full analysis, including photos by what they show and the text in them, or reads a chat's latest messages.
“What did the Pune dealer say about the delayed payment?”
find_person
Finds a person by name, company, Telegram handle or the last digits of their number, with every chat they're in across apps.
“Which chats is Ramesh from Patel Steel in?”
list_chats
Chats with what they're about: category, description and tags. Filter by words, app or connected number.
“Which groups are about quality checks?”

What it can see, and what it can't

Read-only
The assistant can search and read. It can't send messages, change records or change settings.
Your access, nothing more
A token sees one workspace and only the chats and sources its owner can see. If your access changes, so does the token's.
Chats you chose to analyse
Message text comes only from chats set to Full analysis. Chats on Metrics only aren't searchable.
Personal details masked
Phone numbers, email addresses and Telegram handles in messages are masked, and people are returned without their numbers.
Tokens you control
Each token is personal and expires. It stops working when it's revoked or its owner leaves the workspace. Admins can revoke anyone's.
Limits and a log
Up to 120 tool calls a minute per token, and up to 10 tokens per person. Every call is counted, so you can see which tokens are in use.

Questions

Does DataLense run an AI model for these questions?

No. Your assistant does the reasoning on its own account. DataLense only runs the search and the calculation, and returns the results with links to their sources.

Which assistants work?

Any assistant that can connect to a remote MCP server over HTTP and send a header, including Claude Code, Cursor, VS Code with Copilot, Windsurf and JetBrains IDEs.

Can I use it from ChatGPT or Claude on the web?

Not yet. Those apps connect with an OAuth sign-in, and DataLense uses personal tokens today. Use an assistant from the list above for now.

Who can create a token?

Any owner, admin or member of the workspace. Viewers can't. Each person can have up to 10 active tokens, one for each assistant or device, so they can be revoked separately.

Can the assistant be tricked by what's in a message?

DataLense tells the assistant that message text and file contents are data from the business, never instructions. Because access is read-only, a message can't make it change anything in DataLense.

Ask about your business from where you work

Create a workspace, link WhatsApp or your ERP, then add a token under Settings → AI assistants (MCP).

Create your workspace