Skip to content

v0.2.0

Choose a tag to compare

@watson-qg watson-qg released this 26 Oct 01:11
· 49 commits to main since this release
e041768

Upgrading: a tool that rejects your input now returns isError: true instead of a successful
result whose payload carried an error key. Anything reading result["error"] should read the
error flag instead. The messages are unchanged.

MCP specification 2026-07-28 (#8)

  • Tool errors are flagged isError. _err built {"error": message} and nine call sites
    returned it, so on the wire a refusal arrived as a successful result. It raises ToolError
    now, typed NoReturn. The issue reports "2 occurrences"; measured, it is one payload
    construction reached from nine sites.
  • All five tools declare the four annotation hints, sharing one block.
  • The reason for the block is this server's own, not a sibling's. The owner-list servers answer
    openWorldHint: false because their data ships in the wheel; the live-service servers answer
    true because they call somebody's API. This one has no data at all — the moon's position is
    Meeus chapter 47 evaluated at a moment, path loss is the free-space formula. EM-C12 asserts
    that: no data directory, and no file opened at run time.
  • idempotentHint is a stronger claim here than anywhere else in the fleet. An owner-list
    server is idempotent only until the package is upgraded and its data changes underneath. A
    closed-form calculation gives the same answer in 2026 and in 2036. EM-C20 pins a moment and
    checks two calls agree; EM-C21 holds the case that looks like a counterexample and is not —
    when left empty means now, which is the argument defaulting rather than the tool behaving
    differently. It also records plainly that eme_moon_window takes no when and so cannot be
    pinned at all, which is why the stronger claim is demonstrated on the tools that accept a moment.
  • The refusal messages are the useful half, and they survive. A frequency outside 30 MHz to
    25 GHz, a band this server does not name, a time it cannot parse — EM-C31 checks each still
    says what to send instead, including that a refused band names the ones it knows.
  • tests/test_mcp_conformance.py (EM-C01 onward, 14 tests). Every check falsified — 11 of 11.
  • Nothing moved on the wire but the annotations. Every name, description, inputSchema and
    outputSchema dumped through a real fastmcp.Client before and after: byte-identical.

Lint (#7)

  • E501 is adopted. Four lines, not one a defect.
  • Every tool description came out byte-identical, because none of the long lines was in a
    docstring — unlike the owner-list servers, where reflowing a summary line reached a client.
  • The publisher-name exemption the issue was filed for does not apply: this server names no
    publisher's field, because it serves no publisher's list. Zero exemptions; the only noqa is a
    pre-existing F401 on a deliberately planted import in test_security.py.
  • Falsified: a long line planted in src/, caught.

Not applicable here

  • No certifi work and no issue for it (qso-graph-devel#73): there is no outbound call of any
    kind anywhere in the package. EM-C10 keeps it that way and names certifi in its failure
    message.