Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
MEDIMGQ_TS_ROOTNoLocation of the TotalSegmentator dataset../Totalsegmentator_dataset_v201
MEDIMGQ_BLOCKLISTNoPath to the CSV file of cases a reviewer has rejected../qc/blocklist.csv
MEDIMGQ_INDEX_DIRNoLocation where the indexes are stored../data/index
MEDIMGQ_OUTPUT_DIRNoWhere images go when no output_dir is given../outputs

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This server publishes no instructions, or was last inspected before Glama recorded them.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}

Tools

Functions exposed to the LLM to take actions

NameDescription
list_structuresA

List labelable structures. query resolves a name/synonym or substring-searches; filter by region (abdomen, thorax, neck, pelvis, head, spine, upper_limb, lower_limb) or system.

dataset_infoC

Filter values (pathology, study type, contrast phase), windows, presets, and option names.

list_findingsA

Pathology this dataset actually contains, found by measuring the segmentations (aneurysms, goiter, renal cysts, cardiomegaly, organ enlargement), with case counts. Also lists the coarse per-scan pathology categories from the dataset's metadata.

find_findingsB

Cases showing a measured finding (see list_findings), with the axial slice that best shows it. min_value defaults to the finding's threshold. Render a result with render_image(case_id, 'axial', slice, labels=[...structures_to_label]).

find_slicesC

Rank (case, view, slice) candidates where every structure is clearly visible. classical = textbook view and vertebral level; level = a vertebral level like 'L1'.

slice_contentsC

Structures visible on a slice (area, visibility), largest first. Use to choose distractors.

render_imageB

Render a specific slice with labels. Returns image path (+ base64), answer key with arrow/label geometry, and optional hotspot masks (transparent PNGs, pixel-aligned with the image). shuffle_labels randomly reassigns the label texts among the groups (seed to reproduce).

labeled_imageC

Find a slice where every labeled structure (plus context_structures) is visible, then render it. Picks randomly among the top_k best cases so repeated calls vary (fix seed to reproduce). shuffle_labels randomly reassigns the label texts among the groups so the answer's number varies.

batch_labeled_imagesA

Run many labeled_image requests into one folder with a manifest.jsonl (no base64 in the reply). With avoid_repeat_cases each image comes from a different case when possible.

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

B3.2/5.0

Scored across 9 tools

Disambiguation4/5

Each tool has a distinct primary role: metadata, structure listing, finding listing/search, slice search, slice contents, rendering, and batch rendering. The main overlap is that labeled_image is essentially a composite of find_slices + render_image, and batch_labeled_images is a batch variant, but the descriptions make their intended uses reasonably clear.

Naming Consistency3/5

The names mix verb_noun patterns (list_structures, find_findings, render_image) with noun or adjective forms (dataset_info, slice_contents, labeled_image, batch_labeled_images). The convention is still readable, but it is not fully predictable across the set.

Tool Count5/5

Nine tools fit the medical-imaging query/render workflow well, with each tool earning a place and no excessive surface area. The set is large enough to cover search, inspection, rendering, and batching without feeling bloated.

Completeness4/5

The surface covers metadata, structure/finding discovery, slice search, slice inspection, single rendering, and batch rendering, which are the core operations for this domain. Minor gaps exist, such as no explicit case-listing/get_case tool, but agents can work around them through find_* operations.

Maintenance

ActivityMaintained
ResponsivenessNo recent community issues