learning-it by example

#331 AI agents and MCP 6.7 min

Inside an MCP Server: How entity-intel-mcp Works

Ten tools, seven sources, and a gap is never reported as a no.

1. How does the Python demo client talk to the entity-intel-mcp server?
2. In entity-intel-mcp, what does the get_ownership tool return for a Norwegian company (the recorded answer for Equinor)?
3. When the Python demo client for entity-intel-mcp ran with no API keys, how many tools did it list and how many sources were enabled?
4. In entity-intel-mcp, when two sources describe the same company and disagree on a field, how does the merge behave?
5. Why should a stdio MCP server avoid logging to standard output?

Practice only: nothing is saved and no score counts towards anything. Each answer comes with the reason.

  1. Read the whole reason

    Copy the code folder first. Run python show.py norway. Then in show.py change [:78] to [:200] in the norway function and run it again. Say in one sentence what the longer text adds.

    A hint

    The first run cuts the detail line short. The reason is restricted access, not an empty list.

  2. Print when each owner was notified

    In show.py, in the chain function, add a line inside the loop that prints ed['notifiedOn'] and ed['active'] after each 'owns' line. Predict what two values you will see. Run python show.py chain.

    A hint

    Each edge in the recorded data carries these two fields. I saw 2016-04-06 and True for both edges.

  3. List the names of the company

    In show.py, in the entity function, add a loop before the 'legalForm' loop that prints n['kind'] and n['name'] for each n in e['names']. Predict how many names there are and what kinds they have. Run python show.py entity.

    A hint

    The names are in the recorded data under 'names'. I saw one legal name and two former names.

  4. Find the tool that stays offline

    This needs Node 22 and the server file, as in the video. In client.py, in the tools branch, replace the print line inside the loop with: if not a['openWorldHint']: print(t['name']). Predict which tool is printed. Run python client.py tools. No keys or network are used.

    A hint

    The video says only one tool never reaches the network. I saw check_coverage, then the line 10 tools.

These tasks are for you to try on your own machine. Nothing is checked or saved.

What this does not show

Lookups are answers recorded from one live run on 14 November 2026 (public data, one UK group and one Norwegian company); officer lookups and sanctions screening are not shown; the newest MCP spec read has dropped the initialize handshake this server still uses.

Up next in AI agents and MCP

More in AI agents and MCP

All AI agents and MCP videos

Related videos