Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## feature/issue-267-add-mcp #280 +/- ##
=============================================================
+ Coverage 97.68% 98.18% +0.49%
=============================================================
Files 12 12
Lines 1340 1540 +200
Branches 92 146 +54
=============================================================
+ Hits 1309 1512 +203
+ Misses 20 19 -1
+ Partials 11 9 -2
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
|
coverage: 98.766% (+0.3%) from 98.507% — feature/267-mcp-write-payload-reduction into feature/issue-267-add-mcp |
yakky
force-pushed
the
feature/267-mcp-write-payload-reduction
branch
from
October 4, 2026 13:54
6a28cff to
158ec68
Compare
yakky
added this pull request to stack #281
October 4, 2026 13:55
yakky
changed the title
Contributor
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
Payload semantics, membership pagination, batch efficiency, and release-version configuration contain unresolved defects.
Review effort: Balanced
Findings: 4
Open (4)
What changed in this PR
Adds opt-in MCP response reduction for reads and writes while preserving default response behavior.
Changes:
- Adds payload projection, media stripping, field selection, and assignee resolution.
- Adds reduced write representations and batch work-item updates.
- Expands documentation, tests, changelog fragments, and prerelease metadata.
| File | Description |
|---|---|
taiga/mcp_server/server.py |
Adds read/write reduction and batch updates. |
taiga/mcp_server/serialize.py |
Implements response projections and collapsing. |
tests/test_mcp_server.py |
Tests MCP tool behavior. |
tests/test_mcp_server_serialize.py |
Tests serialization helpers. |
docs/mcp.rst |
Documents new parameters and limitations. |
taiga/__init__.py |
Updates package version. |
pyproject.toml |
Updates bumpversion metadata. |
changes/+mcp-write-payload-reduction.feature |
Records write reduction feature. |
changes/+mcp-surface-real-error-messages.bugfix |
Records error-message fix. |
changes/+mcp-resolve-assigned-users-field-name-fix.bugfix |
Records assignee-name fix. |
changes/+mcp-payload-reduction.feature |
Records read reduction feature. |
changes/+mcp-minimal-field-set-fixes.bugfix |
Records minimal-field fixes. |
changes/+mcp-invited-by-collapse.bugfix |
Records membership collapsing fix. |
changes/+mcp-expand-strips-media.bugfix |
Records expanded-media fix. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Neither Milestones.list() nor WikiPages.list() require project at the client/API level (confirmed live against taiga.nephila.it), unlike project-scoped tools such as search or the by_ref lookups. Bring list_milestones and list_wiki_pages in line with the sibling list_user_stories/list_tasks/list_issues/list_epics pattern, which already treat project as optional. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…estone Milestone.user_stories is always fully expanded by python-taiga's parser, making the embedded field potentially large. Add include_user_stories (default True, preserving current output) to both tools so callers can opt into a trimmed response with that key stripped. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…tests Minor code-review follow-ups: replace the Any -> Any signature with the actual dict[str, Any] | list[dict[str, Any]] shape it's always called with, and add unit tests exercising the helper directly (dict, list, and no-op-when-absent cases) rather than only through list_milestones/ get_milestone. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…es toggle Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
A task review found that entity_type/ref/fields were extracted from each item before the try/except block, so a malformed item (missing a required key) raised KeyError outside the per-item handler - aborting the whole batch and discarding results already built for earlier successful items, contradicting the tool's own "not atomic, one row per item" contract. Moves the extraction inside the try block so a malformed item now produces its own error row instead of crashing the batch. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
…prove batch error messages
Addresses final-review findings: `_represent`'s "minimal" branch merged
`{**base, **written}`, letting a caller-supplied stale `version` inside
`fields` (used for the write's own optimistic locking) silently
overwrite the correct post-write version - fixed by reversing the merge
order so `base` always wins. `docs/mcp.rst` claimed `version` is "not a
separate parameter anywhere in this server", which is false -
`set_custom_attribute_value` has one - scoped the claim to
`create_*`/`update_*` tools. `update_work_items`'s error rows used bare
`str(exc)`, producing opaque messages like `"'wiki'"` for structural
failures (unsupported entity_type, missing keys) - now includes the
exception type name.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
…display resolve_assigned_users always returned null names — confirmed against a live Taiga instance that membership records expose the display name as full_name, not full_name_display (the field name every other Taiga user block in this codebase uses, which this code wrongly assumed memberships shared too). The output key stays full_name_display for consistency with those other blocks; only the source field read from the membership record changes. Also corrects an unrelated test's example field name (list_memberships' generic fields-projection test) to use a real membership field instead of the same wrong assumption.
…issue - milestone: add project_extra_info.name/.slug so a cross-project report can display/link each board without a second call (M4) - userstory: add assigned_users so minimal never under-reports secondary assignees (M5) - userstory/issue: drop the redundant per-project status id in favour of status_extra_info.name, which is comparable across boards (M8b) - docs: document verified filter keys and the is_closed/ status_extra_info.is_closed contradiction caveat (M11) Ref: artifacts/specs/2026-09-17-taiga-mcp-fix-recommendations.md Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
…set fixes Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
strict_filters only catches this server's own parameter names nested inside filters by mistake. It gives no protection against a misspelled or unsupported Taiga filter key (e.g. milestone__in, silently ignored by Taiga's REST backend) - that class of mistake returns a normal- looking but wrong result set with no error either way. Document this prominently so filter-based narrowing is re-asserted client-side rather than trusted purely because the call didn't raise. Recommendation 2 (echoing the filters Taiga actually applied) was considered and dropped: Taiga's list API gives no such signal back to this client, so the only thing this server could honestly echo is the query it sent, not what Taiga did with it - materially weaker than what the finding asked for, so it's left undone rather than shipped with an implied guarantee it can't deliver. Ref: artifacts/specs/2026-09-17-taiga-mcp-fix-recommendations.md (M10) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Anticipated failures - a strict_filters violation naming the offending keys, a stale/missing version on a single-item update - were reaching callers as a bare "Error executing tool <name>" with no detail. The mcp SDK replaces any exception that isn't its own ToolError with a fixed generic message, treating it as an unanticipated crash; only ToolError (or a ValueError raised inside a pydantic validator) preserves the real text. Add a _patch() helper wrapping every single-item update_*/update_*_by_id tool's resource.patch() call, re-raising as ToolError with the original message, and change _check_strict_filters to raise ToolError directly. update_work_items already avoided this by catching exceptions itself and returning them as structured rows; single-item tools now surface the same detail. Ref: artifacts/specs/verify-writeside-b5.md, artifacts/specs/taiga-mcp-2.0.0b5-verification.md (N1/N3) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
collapse_extra_info() only matched keys ending in _extra_info, so invited_by (seen on membership records) - a full nested user block - escaped the same shrinking every other user/project block gets, even though it carries the same id + full_name_display shape. Ref: artifacts/specs/verify-readside-b5.md, artifacts/specs/taiga-mcp-2.0.0b5-verification.md (M7) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
apply_payload() previously merged expand's raw content after the strip pass, guaranteeing an expanded block was byte-identical to the raw resource - including avatar/logo fields, which strip_media otherwise removes from everywhere else in the response. Measured as a 2.81x size penalty in practice, with over half of it one duplicated signed logo URL. Reorders the merge before the strip pass instead: expand still adds back every field of the named block that MINIMAL_FIELDS/fields would otherwise drop, but the block now goes through the same strip_media semantics as the rest of the response. This reverses a Day 1 ruling that took the opposite position; usage measurement since then shows the media leak costs more than the completeness guarantee is worth. Ref: artifacts/specs/verify-readside-b5.md, artifacts/specs/taiga-mcp-2.0.0b5-verification.md (M6) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
… issue Two independent live verification sessions found that some MCP clients reject a tool call outright when a parameter carrying a schema-level default is omitted, even though it is absent from the schema's required array - affecting every such parameter, including ones that predate this server's payload-reduction work (e.g. list_milestones' include_user_stories). Confirmed this server's own schema and its own protocol-level validation are correct (live tools/list shows no required entries for these parameters, and a raw stdio JSON-RPC call omitting them succeeds); the rejection traces to the calling client's own schema-to-validator bridge, not to this server, and there is no hook in this server's dependencies to change what gets emitted to work around it. Ref: artifacts/specs/verify-readside-b5.md, artifacts/specs/taiga-mcp-2.0.0b5-verification.md (B1/N1), artifacts/specs/2026-09-17-b5-verification-spike-report.md Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Confirmed via a raw, unfiltered probe: a milestone-embedded user_story has no assigned_users key at all - Taiga's embedded serializer omits it entirely, unlike the top-level list_user_stories/get_user_story shape. Not something fields or resolve_assigned_users can produce, since there's no bare id list in the embedded payload to project or resolve from. The one-call "everything in a sprint" route is real and valuable, but yields primary assignees only. Ref: artifacts/specs/taiga-mcp-2.0.0b5-verification.md (N4) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
…the last id (N2)
filters={"milestone": [1444, 1446]} returns a small, clean, plausible-
looking result set - but only for the last id, with every other id's
items silently dropped. Traced to taiga/requestmaker.py: query is
passed straight to requests.get(..., params=query), and requests
already serializes a list value as repeated same-name query params
correctly - this client is not the one collapsing it. The collapse
happens in Taiga's own REST backend (standard Django QueryDict.get()
semantics on repeated params), outside this repo's control - the only
available mitigation is documenting it, alongside the existing
milestone__in caveat.
Ref: artifacts/specs/taiga-mcp-2.0.0b5-verification.md (N2)
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
Fragments for the three b6 bugfix commits (060e3a4, 4901a62, 40b8f94) that were missed at the time. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QBEycLqZR68HpZF5jCdq7Y
…s resolution Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
yakky
force-pushed
the
feature/267-mcp-write-payload-reduction
branch
from
October 4, 2026 14:32
158ec68 to
903cce4
Compare
…ds is set Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacks on
feature/issue-267-add-mcp(#268). Adds opt-in payload-reduction for the MCP server's read and write tools, plus bugfixes found during live verification against taiga.nephila.it.Read-side (
payload/fields/strip_media/expand/resolve_assigned_users/strict_filterson the full-resourcelist_*/get_*tools (not history or custom-attribute values)) and write-side (return_representationon everycreate_*/update_*tool, plus a newupdate_work_itemsbatch tool) - both opt-in, byte-identical by default.Bugfixes: malformed-batch-item isolation, stale-version merge-order leak,
resolve_assigned_usersfield-name bug, severalMINIMAL_FIELDSgaps, generic error messages now surface the real message,invited_bynow collapses undercompact/minimal,expandnow also strips media.Also documents three issues found with no code fix available: a client-side MCP validation quirk, a Taiga-server-side list-filter collapse, and a genuine Taiga API gap in embedded
user_stories.Replaces #277 (head branch renamed so the towncrier branch-name check passes). The epic-link / version-fill fixes are in #279.
🤖 Generated with Claude Code