Claude, Maltego & OpenAI
An overlapping trio for intelligence work
I have been a Maltego user since 2010 and I became fairly enamored with Claude Desktop once I discovered the Model Content Protocol servers. Earlier in July I was handed an investigation that had some obvious AI prompts in it, but Claude simply refused to touch this stuff. I inquired of the source and learned that OpenAI’s offering is much more open to digging into people, at least when it comes to public figures.
This whole month I’ve woke up, sat at my desk and waded into MCP stuff, grown weary, laid down with my laptop to try other stuff, dozed off, and woke to do it again.
I really feel like I’m on to something, let me explain …
Attention Conservation Notice:
This is me waxing poetical about how cool MCP is and the GraphCraft aspects of making this stuff work with Maltego. I actually got some productivity out of this stuff, I’m officially hooked. If you’re very technical you might build/test/expand some of this yourself, if you’re an analyst there are things in here you will probably want to adopt.
Achievement Unlocked - Productivity:
I don’t know about all of you, but for me I want to quickly get to some sort of productivity gain on anything I’m doing. If I spend an hour learning, try something, and get some time back, I feel good, like I can continue to do this and maybe I’ll get more hits. Prior to Claude Desktop + MCP, AI was a search engine for me. That’s the Boomer/Gen-X mode of use. Everything else I did just had enormous learning curves and little in the way of results.
But with Claude, just in July, I got the following done:
Post surgery I started harassing Claude about my supplement regimen. Once I got the simple local Sqlite3 database running, things just opened up. Now I am tracking everything I do, I got my vitamins/minerals/electrolytes tuned, and it’s legit kicked out some diagnostics that I’m doing to see the doctor about in August.
There are a couple court cases in my queue. I put one of them into mcp-document-server, and instead of reading a War & Peace level of crap, I quickly found the few things I needed to read for my role. This would not have worked with simple text search, it was the vector database/semantics that helped.
Shall We Play A Game? had almost a thousand saved articles in Evernote. About 800 made the jump to Obsidian format, which I can access with mcp-obsidian. I’ve been trying to coax this to load into the new Memento Neo4j memory I’m using, but batch processing with Claude is never not frustrating.
I have this pile of literature to review because its influential in an area that interests me. I fed 4,000 pages into mcp-documentation-server and it kicked out a tolerable summary. I then constrained it to one specific document and got a much more detailed exposition on the area of interest.
I had an RSS thing running, I mentioned it in Situational Awareness Using Claude, but I broke it. Once repaired, I found I needed to divide up alerts a bit to do what I want, but the stable, off net time series of events proved invaluable in attribution.
Future Features:
The mcp-document-server is sweet - tiny, easy to understand, produces results. But it won’t stand up to the Disinfodrome sized loads of some problems. I have Chroma running and Minima is on my radar. Document handling is key to reducing the boundless hallucinations that otherwise plague research efforts.
I am really excited about the Memento memory system, especially if the developer keeps pushing on the compartment by Project angle. I had a couple different Neo4j users, a couple home databases, then one local and one remote Neo4j server, and in general this is just crap. Claude Desktop is just not hip to these enterprise security model requirements. If there’s a slick integration by the native Project mechanism, that’s fantastic. I want to be able to work a case, dump a backup of the memory from it, and hand it off to someone else.
Reasoning is the very next thing I am going to touch. Claude has a “think” option that is different than the built in “extended thinking” it already does, I got this installed, but they’ve been having issues this morning, so haven’t done anything with it yet. There are other inference support tools out there in the form of MCP servers, but I don’t yet have a full inventory, let alone any opinion on what fits. A related thing that looks absolutely fascinating is Chronulus. This is pointedly NOT an MCP extension, but the notion that one can “Forecast and predict anything with Chronulus AI Agents” obviously must be inspected more closely.
OpenAI:
One distressing thing I have learned this week is that OpenAI does much better work on investigating public figures, but there’s no MCP integration till the $100/month level of subscription. They’ve also gone the direction of offering “GPTs” - people build customized environments that embed their expertise. This is interesting, I need to explore further, but it’s pointedly NOT a fit for my own integration urges.
Getting Involved:
Claude MCP has been a technologist’s thing, but I just recently found DXT - Claude Desktop Extensions. This is not cooked yet, but it will be very soon, and you, constant reader, will be able to follow simple formulas and get the same stuff I am using available for your own activities.
Conclusion:
The last time I spent this much time on development I was doing hardcore Twitter streaming at scale in 2019. I’d been running Elasticsearch, I added Redis, then started converting to ArangoDB. The workflow across multiple machines required RabbitMQ. This was, for me “flow”, stuff just kept happening in a rapid, pleasing fashion.
Now for the first time in a good five years, I am there again with this integration stuff I am doing. It will hopefully lead to some paying work exercising it, as well as an opening to build another software service. Three times in the last thirty years I’ve constructed a system that led to roughly three years of gainful employment related to operating it, this feels like another episode starting.
CODA:
This is what Claude had to say for itself when I asked about its MCP configuration
Knowledge & Documentation
Documentation - Document storage and semantic search
Chroma - Vector database for embeddings and similarity search
Obsidian - Note-taking system integration
Think Tool - Structured reasoning and planning
Sequential Thinking - Multi-step problem solving with tool recommendations
Wolfram Alpha - Mathematical and computational queries
Memento - Knowledge graph memory system for persistent memory
SQLite - Database operations for structured data storage
Investigation
Shodan - Internet-connected device discovery and security research
Builtwith - Website technology stack analysis
Sanctions - Compliance and sanctions screening
Web & Research
Web Search - Search the web for current information
Web Fetch - Retrieve content from specific URLs
Fetch - Alternative web content fetching tool
Development & Code
Claude Code - Agentic coding tools (Bash, Read, Write, Edit, etc.)
Analysis Tool (REPL) - JavaScript execution environment for calculations and data analysis
Artifacts - Create and manage code artifacts and documents
Utilities
Time - Time zone conversion and current time
Filesystem - File system operations for reading/writing files
Playwright - Browser automation and web testing