mcpdata Request an invite

Answers / Explainers

Is it safe to let AI query my data?

Recipe verified on Ran on the public demo (read-only)
Short answer

It can be, if access is controlled: give the assistant a key with only the role it needs (Read can't change anything), limit the key to known addresses and an expiry date, and keep an audit log. The data should be encrypted at rest, and the key revocable in one click.

Steps

  1. Connect your assistant to the public demo. It's a read-only instance with 10 million synthetic orders, so you can try this before you have your own. In Claude Code: 
    claude mcp add --transport http mcpdata-demo https://demo.mcpdata.org/mcp --header "Authorization: Bearer cdb_live_PkVoldus8qcE-hqffjajNZzQwiA7rTHMoR4RLSAxsCE"

    The key is public on purpose: it can only read the demo's sample data. For Claude Desktop, Cursor and other apps, see Connect your AI.

  2. Ask your question. Give your AI this prompt:
    Using mcpdata, create a table agent.test with one column x.
What your AI does: the tool call it makes for this prompt, and what came back
create_table(name="agent.test", kind="log", columns=[{"name": "x", "type": "int64"}]) → refused: Error calling tool 'create_table': not authorised: writing tables needs the ReadWrite role -- this agent's API key is read
Ask Claude with this prompt Ask ChatGPT with this prompt

The assistant needs mcpdata connected (step 1) to run the query.

Why this works

The assistant sees only what its key allows, and everything it does is a logged tool call. Below, the public demo's Read key tries to create a table and is refused: a Read key can query, but it can't write, export files, share data or send reports. What the AI provider sees is what your assistant asked for and got back; the tables stay in the database.

Related questions