v0.6.0

Added

• Chain tools (MCPServer.register_chain_tool). Sequence several
  ServiceSpec / SelectorSpec steps behind a single MCP tool, where later
  steps read earlier outputs — retrieve x → write y → write z with z
  derived from both x and y. Each ChainStep binds its result to an
  alias readable via ctx[alias]; an optional inputs(ctx) callable maps
  the validated tool arguments (ctx.args) and prior outputs to that step's
  kwargs. atomic=True (default) runs the whole sequence in one
  transaction.atomic() — any step raising ServiceError /
  ServiceValidationError rolls back every prior write and the JSON-RPC
  error carries failedStep. The advertised inputSchema is the chain's
  explicit input_serializer or the first step's (the first-step fallback);
  the response is the output_alias step (default: last) or {alias: rendered} for every serializer-bearing step under output_all=True. Each
  step's spec.permission_classes are AND-combined with chain-level
  permissions and checked up front. Sync + async. New public exports:
  ChainStep, ChainContext.
• Resolved-data output serializer context (sister-repo 0.15+). Output
  context providers may now declare a keyword for the data about to be
  serialized — result (service tool), instance (selector RETRIEVE), or
  page (selector LIST) — and the value is forwarded through tool dispatch
  (sync + async). This lets a provider run a single batched query against
  the exact objects being rendered instead of re-fetching. view /
  request stay positional, so existing (view, request) providers are
  unaffected. For the LIST path the context is resolved after the page is
  materialized and the provider receives the same object the renderer
  iterates, so an id-keyed batched query reuses the queryset's result
  cache.

Changed

• Bumped djangorestframework-services to ==0.15.0 (additive — the
  resolved-data output-context feature above).

Fixed

• kind=LIST pagination now handles non-QuerySet selector returns. A
  paginated LIST selector that returned a plain list / tuple previously
  hit list.count() (which takes an argument) and failed with the opaque
  count() takes exactly one argument (0 given). Pagination now
  discriminates QuerySet vs sequence with _is_queryset_like: QuerySets
  count via .count(), sequences paginate in-memory via len() + slice,
  and a non-sized/non-sliceable return (e.g. a generator) raises a clear
  "must return a QuerySet or a sized, sliceable sequence" error.