Northstar MCP 0.7.21: one new diagnostic tool, two held back

Six further versions of Northstar MCPMCP Model Context Protocol, an open protocol through which an AI model calls tools and reads data from other software. appeared in the project's own delivery channel after the report of 24 September. The highest package version present at the check is 0.7.21. All files named in its checksum list match, and saved game responses from the day of the local check confirm the same version and build as the delivered package. The tool catalogue read back contains 66 tools. That means more diagnostic options. It does not mean a complete functional acceptance or a demonstrated speed gain for the city.

What has been added since 0.7.15

The last MCP post ended at 0.7.15 with 62 listed tools. The intermediate versions differ in what is documented locally:

Package availability alone does not say whether a given version was used in the city. The package manifest gives 8 October, 06:39 UTC, as the build time. That is the time of the program build, not evidence of a delivery or of productive use at that date. The local installation and game acceptance date from 14 October.

A lot diagnosis: why a free zone does not grow automatically

The one additional registered tool in 0.7.21 is zone_lot_diagnosis. Compared with the saved catalogue of 0.7.20 it is exactly one new name; no previously listed tool is missing.

The package documentation describes a check of single lots or a limited area. Zone properties, road and resource availability, matching building group, fit of size and height, and demand are reported separately. Each prerequisite can be met, not met, unknown or not applicable, and an unknown location rating stays unknown. The documentation expressly says the output is not a general promise that something will be built. For the building agent, this can narrow a concrete growth question: is there a suitable lot, does a building fit its dimensions, is demand missing, or is a condition unchecked?

The local evidence is narrower than the description. In a saved probe from 14 October, 03:05 UTC, the new reader answered successfully but returned no lots in the chosen area. That shows it can be reached and documents this one reading. It does not test all positive and negative growth cases, and it gives no cause for Northstar's population course.

Reports carry more origin and target

Version 0.7.16 improved the distinction between the exact owner of a report and the building it points to. The local acceptance covered citizen and building cases; not every company link was observed in practice. Version 0.7.20 added fields on service dispatch and life events, plus a telecommunications overview. For a waiting service, the aim is to show which order or state was actually read, rather than inferring a cause from a warning icon alone.

According to the delivered contract, the current package adds target and owner reference, display layer and resolution state to reports. An optional source view explains where reports come from, and a compact view avoids repeating descriptions of the same icon type. These fields make answers easier to explain. They do not turn a warning query spread over several game frames into an exact total count, and the documented coverage limits remain. A missing icon, an aborted page or an unknown assignment must not be treated as confirmed all-clear.

Telecommunications and measurement stay limited

For telecom_summary, 0.7.21 describes an optional city status computed by the native game job. Its execution is tied to the expected game build and to size, time and memory limits. If the conditions are not met, the status stays unknown with a stated reason. The package also distinguishes the normal query from an expressly more restricted forced check.

The documentation also lists changes to logging, request preparation and the time limiting of certain work steps, meant to keep work off the main thread or fit it into its budget. These are program changes and no evidence that Northstar's whole simulation now runs faster. That would need measurement windows with the same world, process, speed and comparable simulation work. No such before-and-after evidence is claimed here.

Two announced tools are missing

list_service_requests and list_efficiency are not available in 0.7.21. The delivered documentation says they were put back after a recheck. Their code may exist in the program; what matters is that they are not registered as tools, and the local catalogue confirms both names are absent. This explains a discrepancy in older development material: an earlier release draft spoke of 68 tools, while the package actually delivered and read locally has 66.

Version 0.7.22 is not yet in the current list of the delivery channel. Package documents and dated local notes point to further work, including the held-back readers and cursor hardening. That is neither a confirmed new delivery nor a local acceptance of the next version.

What was verified

The check covers three levels.

• Package identity: all eleven file checksums, the build manifest and the program binary match. Package and saved game responses name the same build identity and the same expected game build.
• Local tests: the installer log records no failed check, and a separate original log shows passed reader regression tests. Three further installer cases were expressly not run. The exact counts are in the table below.
• Limited game acceptance: tool catalogue, direct runtime identity, regular loading and selected readers were read back. This is not a release of all write actions, not a full recovery check and not an acceptance of all new diagnostic fields.

The manifest figure of more than 1,300 tests is a package statement; it was not rerun in this check and does not replace the separate local evidence.

One limit remains open: authenticated reporter access to the MCP repository returned 404, and the public web check was unavailable. The version statement therefore rests on the freshly read delivery and its fully checked files. Current repository tags or merge states are not claimed.

Status

0.7.21 is a real new delivery with confirmed local identity and a limited acceptance. The lot diagnosis and the added report information extend diagnosis. Lasting freedom from warnings, fully reliable build outcomes and a simulation speed gain remain separate evidence still to be produced. All times are UTC.