orca-profiles-mcp
An MCP server that reads and edits OrcaSlicer profiles, expanding the full inheritance chain and tracking the provenance of every value.
README
orca-profiles-mcp
An MCP server that reads and edits OrcaSlicer profiles. It expands the whole
inherits chain and shows which link set every value.
What it does
An OrcaSlicer profile stores only its differences from its parent. Opening the file is not enough to learn the outer wall speed: the value may come from the parent, from the parent's parent, or from the engine's built-in defaults. This server walks the chain to its root, computes the effective values and keeps their provenance.
It handles the parts of Orca that break naive readers:
- parents are resolved by name, across vendors — a Sovol filament inherits from the OrcaFilamentLibrary;
- a missing
Generic …parent falls back toGeneric <material> @System, and renamed profiles are found through theirrenamed_fromlist; - vectors bound to extruder variants merge element by element, including the
nilmarker that means "keep the parent's value here"; - a child may declare more extruders than its parent — a toolchanger on top of a single-extruder machine profile — and both extruders keep their values;
- keys that do not belong to the profile's type are dropped, because Orca drops them too; they are reported instead of being silently applied.
Setup
cd ~/job/orca-profiles-mcp
uv sync
Register with Claude Code:
claude mcp add orca-profiles -- uv --directory ~/job/orca-profiles-mcp run orca-profiles-mcp
Tools
| Tool | Purpose |
|---|---|
get_setup |
discovered roots, Orca version, vendors, selected presets |
list_profiles |
search by type, name, vendor, source |
get_profile |
profile in raw / resolved / traced mode |
get_chain |
inheritance chain to the root |
explain_key |
where a specific value came from |
find_children |
which profiles inherit from this one |
diff_profiles |
compare two profiles |
compare_with_upstream |
compare against the OrcaSlicer repository |
validate |
broken inherits, cycles, unknown keys, redundant deltas |
check_deltas |
verify expansion against the deltas Orca itself wrote |
| reports redundant keys and unexplained vector lengths separately | |
set_values |
set values with delta recomputation |
create_profile |
create a user profile |
rename_profile, delete_profile |
rename and delete |
normalize_profile |
drop keys identical to the parent's |
Values are returned exactly as the files store them: "0.20" stays "0.20",
and "100%" stays a percentage. The engine normalises both when it slices;
this server does not, so what you read is what the profile says and what gets
written back is unchanged.
Writing profiles
Your own profiles are edited freely. Touching a system or bundled one requires
force=true: those files belong to a vendor library, the next profile update
restores them, and every descendant inherits the change.
Writes go through a temporary file and an atomic rename, so an interrupted
write cannot truncate a profile, and a backup copy is made next to the file
first (backup=false disables it). Values are checked before anything is
written — a key belonging to another profile type, a vector where the engine
wants a scalar, or a number where the format requires a string is refused
rather than stored and silently dropped by Orca. Settings your pinned snapshot
does not recognise are carried through untouched, so a profile written by a
newer Orca does not lose them.
OrcaSlicer reads profiles at startup and rewrites them at exit. Edits made while the application is running will be overwritten — close Orca first.
How correctness is checked
Orca's CLI cannot verify inheritance. --export-settings returns the engine
defaults plus the keys present in the files handed to it and never walks
inherits; selecting presets through OrcaSlicer.conf does not activate them;
passing a whole chain as a file list is rejected as a duplicate config file.
All three were tried against 2.4.2.
What Orca does leave behind is evidence of its own expansion: when it saves a
user profile it stores the minimal delta against the fully expanded parent. So
every profile Orca has written is a recorded answer. check_deltas recomputes
those deltas and compares — a mismatch means the resolver expands a parent
differently than the engine did.
This is what tests/test_smoke.py asserts against the installed library, and
it is how the toolchanger vector defect was found.
Keys stored despite already matching the parent are reported separately: they are harmless leftovers from before a parent changed, not resolver errors.
Refreshing the engine snapshot
Engine defaults, the variant key sets and the per-type key lists are captured
from the installed Orca into src/orca_profiles_mcp/data/engine-snapshot.json,
pinned to its version tag. After upgrading OrcaSlicer:
uv run python scripts/build_snapshot.py
uv run pytest
Tests
uv run pytest
Four layers:
- unit tests over a miniature fixture library, one per edge case — cross-vendor
inheritance, the
Genericfallback,renamed_from, cycles, missing parents, stride-1 and stride-2 vectors,nilelements; - write round-trips: read → write unchanged → the file is byte-identical;
- end-to-end tests that spawn the packaged entry point as a subprocess and drive it over MCP stdio, covering tool registration, a full create/edit/verify/delete cycle, and error propagation;
- checks against the installed library, including the delta verification described above.
Tests using the real library skip themselves when no Orca data directory is present; everything else runs against fixtures.
Documents
- Design:
docs/specs/2026-08-17-orca-profiles-mcp-design.md— includes what reading the OrcaSlicer sources established about the profile format, and what the CLI experiments ruled out - Implementation plan:
docs/plans/2026-08-17-orca-profiles-mcp.md
Licence
AGPL-3.0, matching OrcaSlicer itself.
This project contains no OrcaSlicer code, but its inheritance logic is a
line-by-line port of Preset.cpp and PrintConfig.cpp, and the engine snapshot
holds default values and key lists extracted from that source.
Not affiliated with or endorsed by the OrcaSlicer project.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。