v0.1.3

[0.1.3] - 2026-08-02

Fixed

• structlog carried no upper bound, and the index already serves a major past
  the floor. The declared range was structlog>=24.1; PyPI has been serving
  26.1.0. The artefact does not change — the resolver's answer to the next
  fresh install does, and that is exactly how swiss-energy-mcp 0.3.3 became
  uninstallable when mcp 2.0.0 removed the module it imported.

  Now structlog>=24.1,<27. The bound is measured rather than guessed: this package
  installs and imports against structlog 26.1.0 today, so the cap admits what
  demonstrably works and stops only the next, unknown major.

A dependency range only reaches users through a new release, hence the
version bump. No code changed.

v0.1.2

[0.1.2] — 2026-07-30

Fixed

• The User-Agent reports the actual package version again. The published
  0.1.1 sent termdat-mcp/0.1.0 to every upstream — the version string was
  hardcoded and had been left behind by earlier bumps. The version now comes
  from the package metadata, so it can no longer drift from the package.

Documented

• The public API exposes less than the website — a coverage limit of the
  source, not a scope setting on our side. Follow-up to #11: @dfch supplied the
  12 entry IDs the website lists for «Quellensteuer»; the API returns 7 at
  maximum recall (every language, all 11 fields, infix wildcard, all
  classifications and collections), and the two sets overlap in exactly one
  entry, 447912. Fetching the missing IDs directly via /v2/Entry returns
  HTTP 200 with an empty body — they are not served at all, so no query can
  reach them. The one exception, 1557, is served but carries status
  In Bearbeitung in a collection marked «(aufgehoben)», which suggests the
  search index covers validated entries while the website also shows drafts and
  repealed material.

  This corrects an earlier guess. The residual was described in #11 as probably
  a counting difference — the website listing designations where the API counts
  entries. The IDs disprove that: they are twelve distinct entries. Recorded
  here so the wrong explanation does not outlive the measurement.

  Documented in both READMEs, and in the search_terms docstring, where it
  matters most: a model told only that «a term may genuinely be absent» will
  conclude «absent from TERMDAT», which is false for entries the website shows.
  The docstring now distinguishes the two.

Fixed

• Capped mcp at <2. mcp 2.0.0, published 2026-07-28, removed
  mcp.server.fastmcp — the module this server imports. With the previous
  unbounded >=1.28.1 every fresh resolve picked 2.0.0 and failed at import
  with ModuleNotFoundError, in CI and for anyone running pip install alike.
  Verified in both directions: 2.0.0 fails, <2 resolves to 1.29.0 and imports
  cleanly. Migrating to the 2.x API (mcp.server.mcpserver) stays a separate,
  deliberate piece of work.

v0.1.1

[0.1.1] — 2026-07-27

Fixed

• MCP Registry publish was blocked by a missing ownership marker. The
  registry verifies ownership of a PyPI package by looking for an
  mcp-name: <server-name> marker in the published package README. The marker
  was added to README.md after 0.1.0 shipped, so it never reached PyPI — and
  PyPI releases are immutable, so it cannot be added to the 0.1.0 artifact
  retroactively. This release carries it into the package metadata.
• Search was confined to the VARIA classification (#11, reported by
  @dfch). /v2/Search restricts an ID-less query to a "default set (=VARIA)",
  so every unfiltered search covered 1 of 23 subject areas — and reported the
  result as a normal empty answer. search_terms, translate_term and
  check_terms now send the full classification set unless the caller narrows
  it. «Quellensteuer» 0 → 3 entries, «Pensionskasse» 1 → 21.
• fields could widen a search but never narrow it. Unsent Field.* flags
  keep their API-side default (Terminus, Name, Abbreviation,
  Phraseology = true), so fields="Terminus" was a no-op. All eleven flags
  are now sent explicitly.
• Default field set widened to include Definition, Note and Source;
  translate_term and check_terms stay on the four designation fields, so a
  term merely mentioned in a definition is never reported as an equivalent.
• Misleading scope caveat. The search_terms docstring presented an empty
  result as probable out-of-scope, which invited models to invent a designation
  instead of retrying. It now documents Lucene wildcards and asks for a retry;
  an empty SearchResult carries a hint field saying the same.

Security

• SEC-016 (NeighborJack): the SSE transport now binds to 127.0.0.1 by
  default instead of 0.0.0.0. Binding to 0.0.0.0 is an explicit opt-in that
  logs a stderr warning when used outside a container. README/SECURITY updated.
• SEC-021: code-layer egress allow-list (ALLOWED_HOSTS + assert_host_allowed),
  enforced before every request; docs/network-egress.md.
• SEC-018: input bounds at the tool boundary — max_results 1–100, plus
  string/list length limits on search_term, term, terms, entry_ids, fields.
• SEC-007: hardened non-root Dockerfile for SSE deployments.
• SEC-005 / SCALE-002: accepted-risk ADRs for DNS pinning and stateful load
  balancing (docs/adr/0001, 0002).

Added

• Typed configuration via pydantic-settings (settings.py); new env vars
  TERMDAT_MCP_LOG_LEVEL, TERMDAT_MCP_CORS_ORIGINS, TERMDAT_MCP_VOCAB_TTL.
• Structured logging via structlog, pinned to stderr as JSON (logging_config.py).
• FastMCP lifespan owning the shared HTTP client (cleanup on shutdown).
• Explicit CORS for the SSE transport, exposing only Mcp-Session-Id.
• CONTRIBUTING.md / CONTRIBUTING.de.md; scheduled/manual live-test workflow.
• Expanded test suite (12 → 28 offline tests): per-tool coverage, error paths,
  egress allow-list, tool-schema input bounds.
• MCP best-practice audit against the portfolio catalog (68 checks, 36
  applicable) under audits/: production-ready; the 19-item hardening
  backlog from that audit is addressed by the changes above.

Changed

• api_status and the client no longer forward raw upstream exception strings to
  the model (OBS-002 error-detail masking); detail goes to the structured log.
• check_terms runs its per-term lookups concurrently (asyncio.gather) and
  reports progress via ctx when available (ARCH-007 / SDK-003).
• Tools grouped rationale + MCP-primitives note documented in the READMEs.

v0.1.0 — Initial Release

First public release of termdat-mcp — an MCP server for TERMDAT, the
terminology database of the Swiss Federal Administration. Seven read-only
tools expose officially validated designations of Swiss authorities,
departments and legal acts across DE / FR / IT / EN.

Install: uvx termdat-mcp

See CHANGELOG.md for details.