diff --git a/src/test/test_link_preview.py b/src/test/test_link_preview.py
new file mode 100644
index 0000000..5ba3d6b
--- /dev/null
+++ b/src/test/test_link_preview.py
@@ -0,0 +1,152 @@
+"""Unit tests for the link-preview page (get_preview / get_preview_image)."""
+
+import html
+import re
+
+import pytest
+
+from vfbquery import link_preview
+
+
+def _neuron():
+    return {
+        "Id": "VFB_jrchjrch",
+        "Name": "5-HTPLP01_R",
+        "IsIndividual": True,
+        "IsClass": False,
+        "Meta": {
+            "Description": "",
+            "Comment": "tracing status-Roughly traced, cropped-False",
+            "Types": "[adult glutamatergic neuron](FBbt_00058220);[adult serotonergic PLP neuron](FBbt_00110925)",
+        },
+        "Technique": ["Hemibrain-Reconstruction"],
+        "Tags": ["Adult", "Nervous_system"],
+        "Images": {
+            "VFB_00101567": [
+                {"id": "VFB_00101384", "thumbnail": "https://www.virtualflybrain.org/data/VFB/i/jrch/jrch/VFB_00101384/thumbnail.png"},
+                {"id": "VFB_00101385", "thumbnail": "https://x/second.png"},
+            ]
+        },
+    }
+
+
+def _class_with_example():
+    return {
+        "Id": "FBbt_00003748",
+        "Name": "medulla",
+        "IsClass": True,
+        "IsIndividual": False,
+        "Meta": {"Description": "The second optic neuropil, [see](FBbt_00003701) *the* lobula."},
+        "Images": {},
+        "Examples": {"VFB_00017894": [{"id": "VFB_00030624", "thumbnail": "https://x/medulla.png"}]},
+    }
+
+
+# --- id validation -----------------------------------------------------------
+
+@pytest.mark.parametrize("value", ["FBbt_00003748", "VFB_jrchjrch", "VFBexp_FBtp0000001", "GO_0001872"])
+def test_is_term_id_accepts_vfb_shapes(value):
+    assert link_preview.is_term_id(value)
+
+
+@pytest.mark.parametrize("value", ["", None, "medulla", "<script>", "../x", "VFB jrch", "_leading"])
+def test_is_term_id_rejects_free_text(value):
+    assert not link_preview.is_term_id(value)
+
+
+# --- text helpers (mirror pageMetadata.js) -----------------------------------
+
+def test_strip_markdown_keeps_link_labels_and_collapses_space():
+    assert link_preview.strip_markdown("[medulla](FBbt_1) is *big*   and `odd`") == "medulla is big and odd"
+
+
+def test_truncate_breaks_on_a_word_and_adds_ellipsis():
+    out = link_preview.truncate("word " * 100, 50)
+    assert len(out) <= 50
+    assert out.endswith("…")
+    assert not out[:-1].endswith(" ")
+
+
+def test_truncate_leaves_short_text_alone():
+    assert link_preview.truncate("short", 50) == "short"
+
+
+# --- thumbnail selection -----------------------------------------------------
+
+def test_first_thumbnail_prefers_images_over_examples():
+    info = _neuron()
+    info["Examples"] = {"T": [{"thumbnail": "https://x/example.png"}]}
+    assert link_preview.first_thumbnail(info).endswith("VFB_00101384/thumbnail.png")
+
+
+def test_first_thumbnail_falls_back_to_examples_for_a_class():
+    assert link_preview.first_thumbnail(_class_with_example()) == "https://x/medulla.png"
+
+
+def test_first_thumbnail_none_when_nothing_has_one():
+    assert link_preview.first_thumbnail({"Images": {"T": [{"id": "x"}]}, "Examples": {}}) is None
+
+
+# --- title / description -----------------------------------------------------
+
+def test_title_has_name_id_and_site():
+    assert link_preview.build_title(_neuron()) == "5-HTPLP01_R [VFB_jrchjrch] - Virtual Fly Brain"
+
+
+def test_description_for_individual_without_description_uses_types_comment_technique():
+    d = link_preview.build_description(_neuron())
+    assert d.startswith("5-HTPLP01_R is an instance of adult glutamatergic neuron, adult serotonergic PLP neuron.")
+    assert "tracing status-Roughly traced, cropped-False." in d
+    assert "Imaged by Hemibrain-Reconstruction." in d
+    assert d.endswith("for this image on Virtual Fly Brain.")
+
+
+def test_description_for_class_uses_description_only():
+    d = link_preview.build_description(_class_with_example())
+    assert d == "The second optic neuropil, see the lobula."
+
+
+def test_description_never_exceeds_limit():
+    info = _class_with_example()
+    info["Meta"]["Description"] = "long words " * 100
+    assert len(link_preview.build_description(info)) <= link_preview.MAX_DESCRIPTION
+
+
+# --- metadata / page ---------------------------------------------------------
+
+def test_metadata_urls_are_reports_canonical_and_viewer_target():
+    m = link_preview.preview_metadata(_neuron())
+    assert m["url"] == "https://virtualflybrain.org/reports/VFB_jrchjrch"
+    assert m["viewer_url"] == "https://v2.virtualflybrain.org/org.geppetto.frontend/geppetto?id=VFB_jrchjrch"
+    assert m["image"].endswith("VFB_00101384/thumbnail.png")
+
+
+def test_metadata_image_defaults_to_logo():
+    info = _class_with_example()
+    info["Examples"] = {}
+    assert link_preview.preview_metadata(info)["image"] == link_preview.DEFAULT_IMAGE
+
+
+def _tags(page):
+    return dict(re.findall(r'<meta (?:property|name)="([^"]+)" content="([^"]*)"', page))
+
+
+def test_page_carries_every_tag_an_unfurler_reads():
+    page = link_preview.render_preview_html(_neuron())
+    tags = _tags(page)
+    for key in ("og:title", "og:description", "og:url", "og:image", "twitter:card", "twitter:image", "description"):
+        assert key in tags, key
+    assert tags["twitter:card"] == "summary_large_image"
+    assert tags["og:image"] == tags["twitter:image"]
+    assert 'rel="canonical" href="https://virtualflybrain.org/reports/VFB_jrchjrch"' in page
+    assert 'http-equiv="refresh" content="0; url=https://v2.virtualflybrain.org/org.geppetto.frontend/geppetto?id=VFB_jrchjrch"' in page
+
+
+def test_page_escapes_hostile_ontology_text():
+    info = _class_with_example()
+    info["Name"] = 'x"><script>alert(1)</script>'
+    info["Meta"]["Description"] = "a & b < c"
+    page = link_preview.render_preview_html(info)
+    assert "<script>alert" not in page
+    assert html.escape(info["Name"], quote=True) in page
+    assert "a &amp; b &lt; c" in page
diff --git a/src/vfbquery/ha_api.py b/src/vfbquery/ha_api.py
index a74aa91..127b0ad 100644
--- a/src/vfbquery/ha_api.py
+++ b/src/vfbquery/ha_api.py
@@ -61,6 +61,8 @@
 
 import aiohttp
 from aiohttp import web
+
+from vfbquery import link_preview
 import numpy as np
 
 # Pure standard-library module: the /combine expression language, its set
@@ -4182,6 +4184,59 @@ async def handle_catmaid_command(request):
 # Application factory
 # ---------------------------------------------------------------------------
 
+
+async def _term_info_for_preview(request, short_form):
+    """The cached get_term_info result for a preview endpoint, or None.
+
+    Goes through the same L1 cache, coalescer and worker pool as
+    /get_term_info itself, so a preview never costs a second term lookup and
+    a burst of unfurlers for one link collapses to one query.
+    """
+    key = f"term_info:{short_form}"
+    response = await _dispatch_to_pool(request, key, _run_term_info, short_form)
+    if response.status != 200:
+        return None
+    info = json.loads(response.body)
+    return info if info and info.get("Id") else None
+
+
+async def handle_get_preview(request):
+    """GET /get_preview?id=<short_form>
+
+    The term's link-preview page: og:*/twitter:* tags, title, description
+    and thumbnail rendered server-side, with a meta-refresh into the viewer.
+    Meant to be served to unfurlers (Slackbot, Twitterbot, ...) in place of
+    the viewer, which sets those tags only from JavaScript they never run.
+    """
+    short_form = (request.query.get("id") or "").strip()
+    if not link_preview.is_term_id(short_form):
+        return web.Response(text="Error: a VFB term id is required", status=400)
+    info = await _term_info_for_preview(request, short_form)
+    if info is None:
+        return web.Response(text=f"No term found for id={short_form}", status=404)
+    return web.Response(
+        text=link_preview.render_preview_html(info),
+        content_type="text/html",
+        headers={"Cache-Control": "public, max-age=86400"},
+    )
+
+
+async def handle_get_preview_image(request):
+    """GET /get_preview_image?id=<short_form>
+
+    Redirects to the term's thumbnail, or the site logo when it has none:
+    a stable id-keyed image URL for og:image, sitemaps, cards and the MCP.
+    """
+    short_form = (request.query.get("id") or "").strip()
+    if not link_preview.is_term_id(short_form):
+        return web.Response(text="Error: a VFB term id is required", status=400)
+    info = await _term_info_for_preview(request, short_form)
+    if info is None:
+        return web.Response(text=f"No term found for id={short_form}", status=404)
+    target = link_preview.first_thumbnail(info) or link_preview.DEFAULT_IMAGE
+    raise web.HTTPFound(target, headers={"Cache-Control": "public, max-age=86400"})
+
+
 def create_app(max_workers=None, max_concurrent=None, max_queue_depth=None,
                cache_ttl=None, search_concurrency=None, search_cpu_threads=None,
                search_queue_wait=None):
@@ -4238,6 +4293,8 @@ def create_app(max_workers=None, max_concurrent=None, max_queue_depth=None,
     app.router.add_get("/get_known_neurotransmitters", handle_get_known_neurotransmitters)
     app.router.add_get("/get_hierarchy", handle_get_hierarchy)
     app.router.add_get("/get_hierarchy_html", handle_get_hierarchy_html)
+    app.router.add_get("/get_preview", handle_get_preview)
+    app.router.add_get("/get_preview_image", handle_get_preview_image)
 
     # Canonical free-text search (website-equivalent ranking)
     app.router.add_get("/search", handle_search)
diff --git a/src/vfbquery/link_preview.py b/src/vfbquery/link_preview.py
new file mode 100644
index 0000000..ea21cad
--- /dev/null
+++ b/src/vfbquery/link_preview.py
@@ -0,0 +1,148 @@
+"""Link-preview page for a VFB term: the HTML an unfurler sees.
+
+Slack, Bluesky, Teams, Discord and the rest build a link card from the
+``og:*`` / ``twitter:*`` tags in a page's raw HTML; none of them run
+JavaScript. The Geppetto viewer sets those tags per term only from JS, so a
+pasted viewer link unfurls as the site logo. This renders the same tags
+server-side from ``get_term_info`` so a proxy can hand them to an unfurler
+instead -- and a human who lands here anyway is sent on to the viewer.
+
+The title and description logic mirrors ``pageMetadata.js`` in geppetto-vfb
+so a term reads the same wherever its preview is built.
+"""
+
+import html
+import re
+
+SITE_NAME = "Virtual Fly Brain"
+VIEWER_BASE = "https://v2.virtualflybrain.org/org.geppetto.frontend/geppetto"
+REPORTS_BASE = "https://virtualflybrain.org/reports/"
+DEFAULT_IMAGE = "https://www.virtualflybrain.org/favicons/vfb-logo-512.png"
+MAX_DESCRIPTION = 300
+
+_MARKDOWN_LINK = re.compile(r"\[([^\]]*)\]\([^)\s]*\)")
+_ID_PATTERN = re.compile(r"^[A-Za-z][A-Za-z0-9]*_[A-Za-z0-9_]+$")
+
+
+def is_term_id(value):
+    """True for the id shapes VFB uses (FBbt_00003748, VFB_jrchjrch, ...)."""
+    return bool(value) and bool(_ID_PATTERN.match(value))
+
+
+def strip_markdown(text):
+    if not isinstance(text, str):
+        return ""
+    text = _MARKDOWN_LINK.sub(r"\1", text)
+    text = re.sub(r"[*`]", "", text)
+    return re.sub(r"\s+", " ", text).strip()
+
+
+def truncate(text, limit=MAX_DESCRIPTION):
+    if len(text) <= limit:
+        return text
+    cut = text[: limit - 1]
+    space = cut.rfind(" ")
+    return (cut[:space] if space > limit * 0.6 else cut) + "…"
+
+
+def first_thumbnail(info):
+    """First thumbnail across Images then Examples, whatever the template."""
+    for group in (info.get("Images"), info.get("Examples")):
+        if not isinstance(group, dict):
+            continue
+        for images in group.values():
+            if isinstance(images, list) and images and isinstance(images[0], dict):
+                thumbnail = images[0].get("thumbnail")
+                if thumbnail:
+                    return thumbnail
+    return None
+
+
+def kind_of(info):
+    if info.get("IsTemplate"):
+        return "template"
+    if info.get("IsIndividual"):
+        return "image"
+    if info.get("IsClass"):
+        return "class"
+    return "term"
+
+
+def build_title(info):
+    return f"{info.get('Name', '')} [{info.get('Id', '')}] - {SITE_NAME}"
+
+
+def build_description(info):
+    meta = info.get("Meta") or {}
+    name = info.get("Name", "")
+    parts = []
+    description = strip_markdown(meta.get("Description"))
+    if description:
+        parts.append(description)
+    # Types is ';'-separated; ", " (not ",") so the sentence reads as a list.
+    # pageMetadata.js still uses "," here -- worth aligning on its next release.
+    types = strip_markdown((meta.get("Types") or "").replace(";", ", "))
+    if types and info.get("IsIndividual"):
+        parts.append(f"{name} is an instance of {types}.")
+    if not description:
+        comment = strip_markdown(meta.get("Comment"))
+        if comment:
+            parts.append(comment + ".")
+        technique = info.get("Technique")
+        if isinstance(technique, list) and technique:
+            parts.append("Imaged by " + ", ".join(technique) + ".")
+        tags = info.get("Tags")
+        if not parts and isinstance(tags, list) and tags:
+            parts.append(f"{name} (" + ", ".join(tags).replace("_", " ") + ").")
+        parts.append(
+            f"View the 3D image, annotations and queries for this {kind_of(info)} on {SITE_NAME}."
+        )
+    return truncate(" ".join(parts))
+
+
+def preview_metadata(info):
+    """The fields a preview page or card needs, as plain strings."""
+    term_id = info.get("Id", "")
+    return {
+        "id": term_id,
+        "title": build_title(info),
+        "description": build_description(info),
+        "image": first_thumbnail(info) or DEFAULT_IMAGE,
+        "url": REPORTS_BASE + term_id,
+        "viewer_url": f"{VIEWER_BASE}?id={term_id}",
+    }
+
+
+def render_preview_html(info):
+    """A self-contained page carrying the term's link-preview tags.
+
+    Everything an unfurler reads is in the head; the body is a one-line
+    fallback for the odd human, with a meta-refresh into the viewer so they
+    still land where the link pointed. Values are HTML-escaped: names and
+    descriptions come from ontology text and can contain anything.
+    """
+    m = {key: html.escape(str(value), quote=True) for key, value in preview_metadata(info).items()}
+    return f"""<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<title>{m['title']}</title>
+<meta name="description" content="{m['description']}">
+<link rel="canonical" href="{m['url']}">
+<meta property="og:type" content="website">
+<meta property="og:site_name" content="{html.escape(SITE_NAME)}">
+<meta property="og:title" content="{m['title']}">
+<meta property="og:description" content="{m['description']}">
+<meta property="og:url" content="{m['url']}">
+<meta property="og:image" content="{m['image']}">
+<meta name="twitter:card" content="summary_large_image">
+<meta name="twitter:title" content="{m['title']}">
+<meta name="twitter:description" content="{m['description']}">
+<meta name="twitter:image" content="{m['image']}">
+<meta http-equiv="refresh" content="0; url={m['viewer_url']}">
+</head>
+<body>
+<p><a href="{m['viewer_url']}">{m['title']}</a></p>
+</body>
+</html>
+"""