A timeline of "MCP" in my life course.
MCP 4 Era
The first referent of the acronym is the operating system of the same name, which was at release 19 in 2019.
I was the systems programmer at Daytona Beach Community, now Daytona State College which was then a Burroughs shop as my second multi-year job out of college ('83-'85). 
In a my domain space concept, it is the designation for nodes of a Domain Control Program (DCP).
«MCP» is the operating system abstraction on a single node of a cluster, or cloud of computers with fast interconnectivity, miniminally 1 gigabit per second. The MCPs operate as the nodes of the larger OS construct, the DCP. MCP itself has these components/layers:
- The top level which is a distributed lisp image running a generic blackboard model of realtime operations control and knowledge base management.
- The workflow level which is implemented by the Work Flow Language, another Burroughs inspiration, reimagined as a context for literate programming and revival of the job control concept based on an adaptation ofto the DCP context.
- A base layer close to machine level using the c++ actor framework and optionally a custom debian kernel.
So DCP is actually the thing analogous to Unisys MCP, comparing whole OS constructs to each other.
Elliott AI ™
After dropping their CMOS line and completing the transition to commodity Intel, Unisys made MCP available free for personal use to run from Windows 64 bit as "MCP Express". I don't recall learning of this before 2019Q3 and didn bring it up till end of Q1 '20 (version 5). It has to be replaced by a new download every July 31st.
Thus, after 35 years, I now have hands on a current MCP (18) system  and that opens a role for the Unisys MCP in the DCP, where "Unisys MCP" will be used wherever the distinction needs to be made clear. Usages like ODT, DMS II or MARC unambiguously refer to the Unisys product line as we never intended taking more than inspiration from it.Much of the base mainframe stuff is available in the Windows based product and updated for the current epoch, though obviously it's a limited version of the actual priced product that runs on their hardware.
The most natural form of integration of Unisys MCP is to allow it as an alternate to linux node in a DCP. A design principle to inform and guide such an effort is that Unisys MCP shall only have integration with the DCP cognitive architecture and not the physical one, each Unisys MCP in a DCP will be an island (Unisys internetworking notwithstanding) unlike the linux nodes which form a single system image. Thus, the integration can be in software built with the standard Unisys dev kit targeting the DCP cognitive architecture.
Thus, "Elliot AI", where Elliot suggests the Algol, Organick, or perhaps the ET display 🖖. Obviously, the proviso in the README for domain space about intentions versus production realizations applies to this a fortiori. However it is a serious intention and due to the small size of any possible market, I will make anything that does reach the stage of distribution available at no cost (other than standard first class capitation and the passed thru cost of any used hosting resources).
In this period the elements of the DCP are prototyped, marshalled, deployed then productized:
- Get working build of all packages in same form they will ultimately be used in the product.
- Get working build of newly created elements such as the DGUI/SPO and WFL.
- Apply the above to the proto domains.
- Workout in service of the proto domains.
- Do productization/packaging for mass deployment
CP 4721 roughly corresponds to what is produced by 1 and 2 and the AKDOMHST/SVC SKUs to 5.
Sometime between milestone 2 and 4, a MCP shell/remote SPO servicemade available to authenticated users.
Blank for formatting purpose.
- Go There
- At DBCC, the SPO for the 6700 was in my office, it and the production 6800 had between them less than 2 MWord of RAM and there was not much more than 1GB of disk, if that. The MCP XE I run has 2X the processors, 32X the RAM and about 20X the disk. Mill is harder to judge, but i5 3.3GHZ is about 50X times the 150ns B6800 clock, and it does about 83K MIPS vs presumed less than 10 MIPS for the B6xx. The i5 box was bought refurbished for about $100. The B6700 was gotten off a state of FL salvage depot but it was probably used more than a $100 of electricity a month.
- "CANDE", "WFL", and "MCP" are the actual overloads, "SPO" isn't really used in the modern Unisys culture, it's been lost from the mainframe days, so the smalltalk thing I'm doing doesn't map to anything specific, although functionally MARC and the ODT would be the analogs. The original SPO was just an ODT with supervisor permissions. Whose thing, mine or Unisys that is referred to by the overloaded terms will be clear in context (within Unisys MCP or not) and use the same "Unisys" modifier where needed.
- Seems like a good enough place to say I have no interest in the Unisys Univac stuff. There appear to be only a couple hundred MCP sites still running and a good number of them are software houses serving the remainder, a mix of banks, govt units, etc.
- When I first encountered MCP and its arch and product lines they were already decades old and I would remark that if Burroughs had followed the advancement of computer science over that time they would have various features they lacked as common business oriented systems, such as OO, AI, etc. Elliot AI is a possible realization of that, albeit now with 3X times the elapsed time since the inaugural MCP. Current MCP bears mark 59.* which I take it roughly marks years from the first mark release.