AlbumentationsX MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
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
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| render_previewC | Render deterministic preview artifacts for local input images. |
| render_preview_batchC | Render deterministic batch preview artifacts and contact sheets for local input images. |
| search_transformsC | Search AlbumentationsX transform metadata. |
| get_transform_schemaB | Get parameter schema, target support, and summary for one transform. |
| list_feedback_tagsA | List structured feedback tags accepted by adjust_pipeline. |
| list_quality_profilesA | List task-aware quality profiles accepted by preview comparison tools. |
| recommend_recipeC | Recommend a task-aware starter pipeline, quality profile, and preview workflow. |
| validate_pipelineC | Validate a pipeline spec and optional CPU Tensor boundary before export. |
| recommend_pipelineC | Recommend a conservative starter pipeline for a CV task. |
| adjust_pipelineC | Adjust a pipeline from structured preview feedback tags. |
| explain_pipelineC | Explain likely pipeline effects, risks, and useful preview feedback tags. |
| plan_augmentation_policyC | Plan a preview-gated augmentation policy for a task and objective. |
| plan_augmentation_policy_candidatesB | Plan 3-5 preview-gated augmentation policy candidates for side-by-side review. |
| plan_policy_iterationC | Plan the next preview-gated policy iteration from concrete review feedback. |
| export_pipelineC | Export a validated pipeline, optionally as a guarded CPU Tensor Python handoff. |
| plan_dataset_onboardingC | Plan the first safe preview for one local image or an image directory. |
| build_review_packetC | Build one host-facing first-preview handoff for one image or an image directory. |
| run_first_previewC | Run one bounded, validated first preview for a local image or image directory. |
| inspect_dataset_qualityC | Inspect local dataset image quality before first preview rendering. |
| score_dataset_preview_candidatesC | Score several preview candidates as one dataset-level decision set. |
| validate_preview_requestC | Validate a preview request before rendering local preview artifacts. |
| trace_preview_variantB | Read the applied-transform trace for one rendered preview variant. |
| compare_preview_runsC | Compare two preview manifests to guide structured feedback and reproducible tuning. |
| interpret_preview_feedbackC | Convert free-form preview feedback into structured feedback tags. |
| plan_preview_reviewC | Plan the next review action for one baseline-to-candidate preview comparison. |
| rank_preview_candidatesC | Rank multiple candidate preview runs against one baseline. |
| export_preview_reportC | Export a visual preview report with ranking, contact sheets, and decisions. |
| summarize_tuning_sessionC | Summarize a baseline-to-candidate preview tuning step. |
| start_tuning_sessionC | Start a persistent multi-step preview tuning session. |
| record_tuning_session_stepC | Record one candidate comparison inside an interactive tuning session. |
| list_tuning_sessionsC | List persisted interactive preview tuning sessions. |
| export_tuning_sessionB | Export one interactive tuning session as Markdown or JSON. |
| close_tuning_sessionC | Close an interactive tuning session as accepted or rejected. |
| archive_tuning_sessionB | Archive an interactive tuning session without deleting its audit trail. |
| cleanup_tuning_sessionsB | Delete older interactive tuning sessions, protecting active sessions by default. |
| record_preview_feedbackC | Persist user feedback for one concrete preview image variant. |
| list_preview_feedbackC | List concrete preview feedback records. |
| record_tuning_decisionC | Persist a local tuning decision for one preview comparison. |
| list_tuning_decisionsC | List persisted local tuning decisions. |
| export_tuning_reportC | Export persisted tuning decisions as markdown or JSON. |
| list_preview_runsC | List recent preview runs recorded under the configured artifact root. |
| get_preview_manifestC | Return the manifest JSON for one recorded preview run. |
| delete_preview_runA | Delete one preview run and its artifacts from the configured artifact root. |
| cleanup_preview_runsC | Delete older preview runs beyond a retention count. |
| diagnose_environmentC | Diagnose local MCP setup, root access, artifact writes, and public surface discovery. |
| run_host_smoke_checkC | Run a read-only host preflight; reading the client-smoke resource is optional. |
| get_workflow_exampleB | Return an active-profile workflow example when the MCP host cannot read resources. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| build_robustness_augmentation_session | Guide an assistant through preview-driven augmentation tuning. |
| compare_preview_runs_for_feedback | Guide an assistant through preview run comparison before adjustment. |
| run_first_preview_review | Guide an assistant through a first local preview with request validation. |
| tune_pipeline_from_preview_feedback | Guide an assistant through structured preview feedback adjustment. |
| export_reproducible_pipeline | Guide final reproducible pipeline export after preview acceptance. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| AlbumentationsX Preview Review | Interactive review surface for rendered AlbumentationsX preview batches. |
| transforms_catalog | Return the transform catalog as compact JSON. |
| pipeline_schema | Return the JSON schema for pipeline specs. |
| feedback_tags_resource | Return structured feedback tags accepted by adjustment tools. |
| quality_profiles_resource | Return task-aware quality profiles accepted by comparison tools. |
| recipes_catalog_resource | Return task-aware recipe recommendations as compact JSON. |
| policy_assistant_contract_resource | Return the preview-gated policy assistant safety contract. |
| capabilities_resource | Return operational limits and safety boundaries for this MCP server. |
| diagnostics_guide_resource | Return the MCP host diagnostics playbook. |
| workflows_catalog | Return built-in agent workflow guides as compact JSON. |
| task_profiles_resource | Return task-specific workflow profiles as compact JSON. |
| preview_tuning_workflow | Return the preview-driven augmentation tuning workflow guide. |
| annotation_preview_workflow | Return the annotation-aware preview workflow guide. |
| client_smoke_example | Return the MCP host smoke-check example. |
| first_preview_example | Return the MCP first local preview host example. |
| distortion_review_example | Return the MCP distorted robustness review example. |
| dataset_onboarding_example | Return the MCP dataset onboarding host example. |
| diagnostics_example | Return the MCP host diagnostics example. |
| review_loop_example | Return the concrete preview feedback host example. |
| report_handoff_example | Return the visual report handoff host example. |
| torch_cpu_compose_example | Return the CPU Tensor Compose validation and export example. |
TDQS
Scored across 47 tools
Many tools have overlapping purposes, such as render_preview vs run_first_preview, recommend_recipe vs recommend_pipeline, and plan_augmentation_policy vs plan_augmentation_policy_candidates. While descriptions help, the boundaries between several tools are unclear, leading to potential misselection.
All tool names follow a consistent snake_case verb_noun pattern, which is a solid structure. However, the variety of verbs for similar actions (e.g., render vs run, recommend vs plan) and some compound names introduce minor inconsistencies in predictability.
With 47 tools, the server exceeds the 'too many' threshold (25+). The scope could be covered by a smaller, more focused set of tools, and the excessive count suggests over-engineering.
The tool surface covers the full workflow: pipeline recommendation, validation, export, preview rendering, feedback collection, tuning sessions, decisions, and environment diagnostics. No obvious gaps exist, though the redundancy in tools masks some minor missing operations like direct update/delete for certain entities.