diff --git a/mdformat_myst/plugin.py b/mdformat_myst/plugin.py
index 88f2a1a..c12de21 100644
--- a/mdformat_myst/plugin.py
+++ b/mdformat_myst/plugin.py
@@ -2,10 +2,13 @@
 
 import re
 import textwrap
+from typing import Dict
 
 from markdown_it import MarkdownIt
+from markdown_it.rules_core.state_core import StateCore
 import mdformat.plugins
 from mdformat.renderer import RenderContext, RenderTreeNode
+from mdit_py_plugins.attrs import attrs_block_plugin
 from mdit_py_plugins.dollarmath import dollarmath_plugin
 from mdit_py_plugins.myst_blocks import myst_block_plugin
 from mdit_py_plugins.myst_role import myst_role_plugin
@@ -45,6 +48,9 @@ def update_mdit(mdit: MarkdownIt) -> None:
     # Enable dollarmath markdown-it extension
     mdit.use(dollarmath_plugin)
 
+    # Enable support for attribute tagging for paragraphs and other "blocks"
+    mdit.use(attrs_block_plugin)
+
     # Trick `mdformat`s AST validation by removing HTML rendering of code
     # blocks and fences. Directives are parsed as code fences and we
     # modify them in ways that don't break MyST AST but do break
@@ -52,6 +58,70 @@ def update_mdit(mdit: MarkdownIt) -> None:
     mdit.add_render_rule("fence", render_fence_html)
     mdit.add_render_rule("code_block", render_fence_html)
 
+    # Force `mdformat` to treat "equivalent" attribute sets in a given HTML element
+    # (e.g., `<p id="a" key1="value1">` as equivalent to `<p key1="value1" id="a">` by
+    # just sorting all such key/value groups.
+    #
+    # Multiple block attributes that are stacked on top of each other can create output
+    # HTML attribute orderings (from mdit_py_plugins.attrs's parsing logic) that cannot
+    # be replicated if we (nicely, for a formatter) collapse those blocks into a single
+    # nicely-ordered attr dict. Therefore, there is no way to avoid doing this sorting,
+    # i.e., we cannot just "preserve" the input ordering.
+    mdit.core.ruler.push("sort_attrs", _sort_attrs)
+
+
+def _sort_attrs(state: StateCore) -> None:
+    """Sort attributes in all tokens to ensure deterministic HTML rendering.
+
+    This fixes validation errors where `mdformat` thinks the HTML has changed
+    simply because the attribute order flipped (e.g. `id="..." class="..."`
+    vs `class="..." id="..."`).
+    """
+    for token in state.tokens:
+        if token.attrs:
+            token.attrs = dict(sorted(token.attrs.items()))
+
+
+def _reconstruct_attrs(attrs: Dict[str, str | int | float]) -> str:
+    if not attrs:
+        return ""
+
+    parts = []
+    if "id" in attrs:
+        parts.append(f"#{attrs['id']}")
+    if "class" in attrs:
+        assert isinstance(attrs["class"], str), (
+            "mdit_py_plugins.attrs guarantees a string here."
+        )
+        for cls in attrs["class"].split():
+            parts.append(f".{cls}")
+    for k, v in attrs.items():
+        if k in {"id", "class"}:
+            continue
+        parts.append(f'{k}="{v}"')
+
+    if not parts:
+        return ""
+    return "{" + " ".join(parts) + "}"
+
+
+def _append_attrs_postprocessor(
+    text: str, node: RenderTreeNode, context: RenderContext
+) -> str:
+    """Prepend MyST attributes to the already-rendered text."""
+    attrs_str = _reconstruct_attrs(node.attrs)
+    if attrs_str:
+        return f"{attrs_str}\n{text}"
+    return text
+
+
+def _paragraph_postprocessor(
+    text: str, node: RenderTreeNode, context: RenderContext
+) -> str:
+    """Encapsulate all paragraph post-processing."""
+    text = _escape_paragraph(text, node, context)
+    return _append_attrs_postprocessor(text, node, context)
+
 
 def _role_renderer(node: RenderTreeNode, context: RenderContext) -> str:
     role_name = "{" + node.meta["name"] + "}"
@@ -117,7 +187,6 @@ def _escape_paragraph(text: str, node: RenderTreeNode, context: RenderContext) -
     lines = text.split("\n")
 
     for i in range(len(lines)):
-
         # Three or more "+" chars are interpreted as a block break. Escape them.
         space_removed = lines[i].replace(" ", "")
         if space_removed.startswith("+++"):
@@ -155,4 +224,14 @@ def _escape_text(text: str, node: RenderTreeNode, context: RenderContext) -> str
     "math_block": _math_block_renderer,
     "fence": fence,
 }
-POSTPROCESSORS = {"paragraph": _escape_paragraph, "text": _escape_text}
+POSTPROCESSORS = {
+    "blockquote": _append_attrs_postprocessor,
+    "colon_fence": _append_attrs_postprocessor,
+    "fence": _append_attrs_postprocessor,
+    "heading": _append_attrs_postprocessor,
+    "table": _append_attrs_postprocessor,
+    # Paragraphs require special handling to escape strings like "++", but also need to
+    # be able to have attrs added.
+    "paragraph": _paragraph_postprocessor,
+    "text": _escape_text,
+}
diff --git a/tests/data/fixtures.md b/tests/data/fixtures.md
index 0b70e44..505df47 100644
--- a/tests/data/fixtures.md
+++ b/tests/data/fixtures.md
@@ -137,6 +137,40 @@ That's a myst target^
 The escape has no effect
 .
 
+Document block attribute order stability
+.
+{ #id2    key1="value1" .class1 .class2  key2="value2" }
+block
+.
+{#id2 .class1 .class2 key1="value1" key2="value2"}
+block
+.
+
+Block attribute collapsing
+.
+ {#id1   .class1  key1="value1"}
+{#id2 .class2 key2="value2"}
+block
+.
+{#id2 .class1 .class2 key1="value1" key2="value2"}
+block
+.
+
+Attribute attached to flowchart
+.
+{caption="`GROUPS` - Description."}
+```mermaid
+flowchart LR
+    id
+```
+.
+{caption="`GROUPS` - Description."}
+```mermaid
+flowchart LR
+    id
+```
+.
+
 Dollarmath inline
 .
 Inline math: $a=1$