v0.2.1

[0.2.1] - 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.

v0.2.0

[0.2.0] — 2026-08-02

This release exists so that a repair reaches the people running the server:
the published 0.1.0 cannot be installed any more. It declares mcp with
no upper bound, and mcp 2.0.0 removed mcp.server.fastmcp — so a fresh
pip install lindas-mcp resolves to 2.0.0 and the console script dies on
startup with ModuleNotFoundError. Measured against the real artefact in an
empty venv, cold and warm interpreter alike.

The repository has carried the fix since the 2.x migration was merged; it was
simply never released, and main kept the same version number as the broken
artefact — so nothing contradicted it.

Changed (breaking)

• Migrated to the mcp Python SDK 2.x. The server API moved from
  mcp.server.fastmcp to mcp.server.mcpserver with no compatibility shim,
  and the dependency is now mcp>=2.0.0,<3. The tool surface is unchanged —
  what breaks is embedding this server's Python API and the dependency floor.
  Anyone who must stay on mcp 1.x should stay on 0.1.0, and pin an upper
  bound themselves, because the published 0.1.0 has none.

Added

• Portfolio-standard repository structure: CONTRIBUTING.md/.de,
  SECURITY.md/.de, EXAMPLES.md, PUBLISHING.md, docs/network-egress.md,
  docs/roadmap.md, Dockerfile, compose.yaml, .dockerignore, .gitignore,
  claude_desktop_config.json, server.json (MCP Registry metadata), and
  .github/ workflows (ci, live, publish) + dependabot.yml.
• tool-definitions.lock.json with a SEC-022 CI integrity check: a committed
  hash snapshot of every tool name and its argument surface, verified by
  server.tool_manifest() so a silent rug-pull fails the build.
• Full mcp-audit run under audits/2026-07-26T125407-Z-lindas-mcp/
  (production-ready; 11 hardening findings).
• Audit remediation — structured logging (OBS-003): stderr-bound JSON logs
  via structlog (logging_config.py), per-tool-call events (tool, duration,
  outcome) and a LOG_LEVEL env var. stdout stays reserved for JSON-RPC.
• Audit remediation — not-found heuristics (ARCH-003): search_cubes and
  resolve_municipality now return match_type (exact/none) and an
  actionable suggestion on an empty result.
• Audit remediation — Context injection (SDK-003): tools accept a Context
  and emit progress/debug events for long-running SPARQL calls.

Changed

• Egress hardening (SEC-021): added a code-layer ALLOWED_HOSTS allow-list
  and assert_host_allowed() in lindas/client.py; the HTTP client now uses
  follow_redirects=False so an off-host redirect is surfaced as an error.
• Binding default (SEC-016): the SSE / streamable-http transport now defaults
  HOST to 127.0.0.1 (loopback) instead of 0.0.0.0; binding to all
  interfaces is an explicit opt-in and prints a stderr warning. The container
  image sets HOST=0.0.0.0 deliberately.
• Connection pooling (SDK-001): a single httpx client is now built once by
  a FastMCP lifespan and shared across all tool calls (client_session()),
  instead of a fresh client per call. Direct unit-test calls fall back to a
  per-call client.
• Tool annotations (ARCH-009): all tools now also set idempotentHint: true
  and openWorldHint: true.
• Schema-level input validation (SEC-018): tool arguments use
  Annotated[..., Field(ge=/le=/min_length=/max_length=)] — out-of-range inputs
  are rejected as a ValidationError at the boundary instead of being clamped.
• Error masking (OBS-002): unexpected exceptions are logged server-side and
  surfaced to the LLM as a generic message; SparqlError/UpstreamError (which
  carry only the public endpoint's own diagnostics) still propagate unchanged.
• CORS (SDK-004): the HTTP transport serves a CORS-wrapped app exposing the
  Mcp-Session-Id header, with origins configurable via ALLOWED_ORIGINS
  (comma-separated; default *) instead of a hardcoded wildcard.
• Documented the single-file server.py rationale (ARCH-011) and the MCP
  protocol-version policy (ARCH-012).
• Aligned pyproject.toml to the portfolio floor (mcp>=1.28.1, structlog,
  Python 3.13 classifier, Issues URL).

v0.1.0 — Initial Release

[0.1.0] — 2026-07-21

Added

• Initial release. 7 read-only tools over the LINDAS SPARQL endpoint:
  search_cubes, get_cube_structure, query_cube_observations,
  list_publishers, resolve_municipality, run_sparql, api_status.
• Three-layer, extraction-ready lindas/ package: client.py (raw SPARQL/HTTP),
  queries.py (anchored templates), cube.py (vocabulary guardrail).
• Two-phase cube access with automatic code-to-label resolution.
• Version deduplication in search_cubes (latest_only, default on).
• Dual transport via LINDAS_MCP_TRANSPORT; retry 2s/4s/8s; 400 passthrough.
• Bilingual documentation (EN/DE) and full probe report in docs/probe-lindas.md.

Known findings

Discovered during the live probe on 2026-07-21.

• Broad SPARQL times out. COUNT(*) over the whole store and DISTINCT ?g
  over all named graphs ran 70–90 s into a timeout (HTTP 000). The same question
  anchored on ?x a cube:Cube answers in ~2 s. Every template is anchored;
  run_sparql warns and caps runtime.
• Observations hang off cube:observationSet, never directly off the cube.
  The naive ?cube cube:observation ?obs returns zero rows. Caught by a live
  test, not by mocks.
• Dimension values are codes, not labels (region 1805, level 4). Code
  lists resolve via sh:in → rdf:rest*/rdf:first → schema:name. Cantons are
  coded as ld.admin.ch/canton/<n> even without an sh:in list; the bare number
  is surfaced.
• Cubes are versioned in the URI (.../cube/2024-1) with schema:version
  and schema:creativeWorkStatus. Search deduplicates to the newest published.
• Licences are per cube and often Fedlex URIs (fedlex.data.admin.ch/eli/cc/),
  which doubles as a join to fedlex-mcp.
• Publishers deduplicate cleanly on dcterms:creator (a single URI), not on
  schema:publisher (multilingual, splits one body into several rows).