Skip to content

Release 1.14.0 - #37

Merged
theodor90 merged 1 commit into
mainfrom
release-1.14.0
Aug 25, 2026
Merged

theodor90 merged 1 commit into
mainfrom
release-1.14.0

Conversation

@theodor90

Copy link
Copy Markdown
Owner

Tool count 35 → 36, free tools 22 → 23, gated unchanged at 13.

Adds a tool, so this is a minor rather than a patch.

Added — get_insider_directory

Browse insiders alphabetically by surname: the A–Z rail with a count per letter, plus one page under the requested letter. Free. Wraps GET /v1/insiders/directory, which shipped in the backend as insiderapi #238 and had no MCP tool until now.

It returns one row per filer group — a fund group files a single Form 4 listing several reporting owners (the fund, its GP, its management company), each a real EDGAR filer with its own CIK, and listing all of them spent about 11% of a capped surface describing the same actors more than once.

It surfaced from codegen:check failing on main with a populated diff body, which is what distinguishes real drift from the documented Windows CRLF artefact. That's the second release running where a missing tool was found by an unrelated codegen run rather than by anything watching the spec.

Changed — the reason to upgrade

List results render as a pipe table instead of pretty-printed JSON. A 100-row get_transactions response goes from ~2,100 lines and ~15k tokens to 103 lines — 60,302 → 22,145 characters, 2.7× — because roughly 60% of the old payload was the same nineteen key names repeated a hundred times. That saving comes out of the caller's context window, not ours.

Two limits are deliberate: no columns are dropped, and rows carrying populated nested collections keep their old JSON exactly (get_signals returns each signal with its transactions[] attached, and those are the substance of the answer).

Fixed — a regression caught before it shipped

formatCell rounded every non-integer to 2dp, on the reasoning that money reads fine that way. It does. Returns don't — they're stored as fractions, and the generated spec says so on every scored endpoint:

field true at 2dp read as
avgReturn3m 0.0234 0.02 +2%, not +2.34%
avgReturn3m −0.003 -0.00 nonsense

GET /v1/insiders/leaderboard returns exactly those fields as flat rows, so it takes the new table path and its whole ranking column would have collapsed into ties.

Release checklist

Followed the one at the top of CHANGELOG.md, including the step the skill docs omit: server.json carries its own version and the MCP registry reads that, not package.json — so both its top-level version and packages[0].version are bumped, along with the tool count in its description. Skipping it fails the registry publish with cannot publish duplicate version.

Also corrects a stale 35 tools registered line in the test output that #36 missed.

Verification

Build clean, tsc --noEmit clean, 131 assertions pass. Version is consistent across package.json, server.json and server.json.packages[0]. Tarball inspected before publish: 80 files, 65.1 kB packed, contains dist/format.js and the new tool at /v1/insiders/directory, and returns precision verified preserved in the built output.

Adds a tool, so this is a minor rather than a patch: get_insider_directory
wraps GET /v1/insiders/directory, which shipped in the backend as insiderapi
#238 and had no MCP tool until now. It surfaced from codegen:check failing on
main with a populated diff body — the second release running where a missing
tool was found by an unrelated codegen run rather than by anything watching
the spec.

The change worth upgrading for is the output format. List results render as a
pipe table instead of pretty-printed JSON: a 100-row get_transactions response
goes from ~2,100 lines and ~15k tokens to 103 lines, 2.7x smaller, because
roughly 60% of the old payload was nineteen key names repeated a hundred
times. That saving is spent from the caller's context window, not ours.

It also carries a fix that would have shipped as a regression. formatCell
rounded every non-integer to two decimals on the reasoning that money reads
fine that way. Returns do not: they are stored as fractions, so 0.0234 printed
as 0.02 — +2% rather than +2.34% — and a -0.30% return printed "-0.00".
/v1/insiders/leaderboard returns exactly those fields as flat rows, so its
whole ranking column would have collapsed into ties.

Tool count 35 -> 36, free tools 22 -> 23, gated unchanged at 13.

Follows the release checklist at the top of CHANGELOG.md: server.json carries
its own version and the MCP registry reads THAT rather than package.json, so
both its top-level version and packages[0].version are bumped along with the
tool count in its description.
@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

@theodor90
theodor90 merged commit 5f00602 into main Aug 25, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant