Atlas / MCP servers / ibm / AssetOpsBench

AssetOpsBenchBLOCK

Name: AssetOpsBench
Author: ibm

mcp/ibm/assetopsbench

MCP server by ibm · ai for physical assets · condition based maintenance · hvac maintenance · Audited 2026-09-28

AssetOpsBench - Industry 4.0: A unified benchmark and framework for building, orchestrating, and evaluating domain-specific AI agents for Industry 4.0 asset operations and maintenance, with 460+ scenarios, 5 specialist agents (IoT, FMSR, TSFM, Work Order,...), and multi-agent orchestration blueprint

Verdict

BLOCK

Grade

Trust score

69 /100

Exposed tools

69 58r · 11w · 0d

Transport

stdio

License

Apache-2.0

Stars

2,312

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

AI Agents for Industrial Asset Operations & Maintenance

A unified, open framework for building, orchestrating, and evaluating domain-specific AI agents in Industry 4.0.

[](https://github.com/IBM/AssetOpsBench/stargazers) [](https://github.com/IBM/AssetOpsBench/network/members) [](LICENSE) [](#publications) [](#ai-competitions) [](#ai-competitions)

[](#) [](#) [](#publications) [](#publications) [](#publications) [](#publications) [](#publications) [](#publications)

📄 **Paper** · 🤗 **Dataset** · 🎮 **Playground** · 📢 **IBM Blog** · 🎥 [Video](https://www.youtube.com/watch?v=kXmB

Read from source at commit 76c35330ddcbOBSERVED · 2026-09-28

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add assetopsbench-mcp --env COUCHDB_PASSWORD=${COUCHDB_PASSWORD} --env LITELLM_API_KEY=${LITELLM_API_KEY} --env WATSONX_APIKEY=${WATSONX_APIKEY} -- uvx assetopsbench-mcp

claude-desktop

{
  "mcpServers": {
    "assetopsbench-mcp": {
      "command": "uvx",
      "args": [
        "assetopsbench-mcp"
      ],
      "env": {
        "COUCHDB_PASSWORD": "${COUCHDB_PASSWORD}",
        "LITELLM_API_KEY": "${LITELLM_API_KEY}",
        "WATSONX_APIKEY": "${WATSONX_APIKEY}"
      }
    }
  }
}

Exposed tools (69)

58 read · 11 write · 0 destructive.

Tool	Risk	Description
`add_failure_modes`	write	WRITE failure modes for an asset class to the database.
`assess_vibration_severity`	read	Classify vibration severity per ISO 10816.
`asset_detail`	read	Return registry details for one asset.
`asset_ids`	read	List asset identifiers registered at one site.
`assets`	read	List assets at one site with compact registry metadata.
`calculate_bearing_frequencies`	read	Compute bearing characteristic frequencies (BPFO, BPFI, BSF, FTF) from geometry.
`compute_envelope_spectrum`	read	Compute the envelope spectrum for bearing fault detection.
`compute_fft_spectrum`	read	Compute FFT amplitude spectrum of a stored vibration signal.
`count_features`	read	Count the feature catalog cards by kind.
`count_models`	read	Count the models in the catalog.
`current_date_time`	read	Provides the current date time as a JSON object.
`current_time_english`	read	Returns the current time in English text.
`data_quality`	read	Assess data quality for a time-series dataset and produce a cleaned file pointer.
`deprecate_feature`	read	Mark a feature catalog card as deprecated.
`deprecate_model`	read	Retire a model card by setting `status=deprecated`.
`describe_candidates`	write	Return a shortlist of candidate models for a task, in catalog order.
`describe_features`	read	Describe specific feature cards by name.
`describe_models`	read	Return a compact record for each of the given model ids.
`diagnose_vibration`	read	Full automated vibration diagnosis pipeline.
`extract_features`	read	Compute scalar feature values from a series with the named extractors.
`find_assets_by_sensors`	read	Find site assets by installed or measured sensor names.
`find_models`	read	Filter the model catalog for a task and return a ranked shortlist.
`generate_failure_modes`	read	GENERATE a new or extended failure-mode list for an asset class.
`get_asset_catalog`	read	Return cataloged asset classes and categories.
`get_failure_mode_catalog`	read	Return cataloged failure modes by asset category.
`get_failure_modes`	read	READ the known failure modes for an asset class.
`get_feature`	read	Return one feature catalog card by feature id.
`get_feature_lineage`	read	Return the parent and descendant chain for a feature catalog card.
`get_model_lineage`	read	Return a model card
`get_result`	read	Fetch one persisted result by task type and result id.
`get_run`	write	Fetch one run record by id.
`get_sensor_catalog`	read	Return cataloged sensor types.
`get_vibration_data`	read	Fetch vibration sensor time-series from CouchDB and load into the analysis store.
`hf_stats`	read	Look up HuggingFace popularity for a model card or repo.
`history`	read	Return one chronological page of telemetry observations for an asset.
`installed_sensors`	read	List sensor names assigned to an asset in the registry.
`json_reader`	read	Reads a JSON file, parses its content, and returns the parsed data.
`latest_reading`	read	Return the newest telemetry observation for an asset.
`list_domains`	read	List the distinct domains present in the model catalog, with counts.
`list_features`	read	List feature catalog cards from the configured database.
`list_known_bearings`	read	List all bearings in the built-in database with their geometric parameters.
`list_models`	read	List model cards in the catalog, optionally filtered by task / domain.
`list_results`	read	List persisted results for a task type, optionally narrowed by asset or scenario.
`list_runs`	write	List run records and plans, optionally filtered by asset.
`list_tasks`	read	List the standardized TSFM tasks available in the benchmark.
`list_vibration_sensors`	read	List available sensor fields for an asset.
`measured_sensors`	read	List measurement fields observed in an asset
`model_template`	read	Return the template for authoring a new model card.
`new_feature_version`	write	Create a successor version for a transform feature.
`new_model_version`	write	Create a successor version of a model card.
`profile_series`	read	Profile a time-series dataset behind a file pointer.
`recipe_template`	read	Return the template for authoring a recipe for run_recipe / run_tabular_recipe.
`register_feature`	read	Register an executable transform feature card.
`register_finetuned`	read	Register a fine-tuned model as a card pointing at its checkpoint.
`register_model`	read	Register a model card in the catalog.
`resolve_model`	read	Preflight a model card: check that it can actually be loaded.
`run_plan`	write	Execute a plan: a DAG of recipes chained by file pointers.
`run_recipe`	write	Run a forecasting or anomaly-detection recipe on a target series from a file pointer.
`run_tabular_recipe`	write	Run a series-to-tabular recipe: regression, classification, or clustering.
`search_features`	read	Search feature catalog cards by id, name, description, or tags.
`search_models`	read	Substring (case-insensitive) search over the model catalog.
`select_features`	read	Rank candidate extractors on one series and return the shortlist worth keeping.
`sensor_coverage`	read	Summarize non-null observation coverage for every measured sensor.
`sensor_stats`	read	Compute numeric statistics for one or all measured sensors.
`sites`	read	List sorted site identifiers available in the asset registry.
`stream_extent`	read	Inspect the size and timestamp span of an asset
`update_feature`	write	Patch fields on an existing feature catalog card.
`update_model`	write	Patch fields on an existing model card.
`wo__list_workorders`	read	Test MCP tool

Trust audit

BLOCKgrade D · trust 69/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	PASS
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: none-observed
Network: declared (8 observation(s))
Shell: declared (3 observation(s))
Dependencies: pinned
Secrets in source: none-found

Findings (17)

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/servers/tsfm/engine/feature_runner.py:27

exec(code, ns)  # noqa: S102 - intentional; sandbox in production

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

MEDIUMCode injection · code.dynamic_import · CWE-78, CWE-94, CWE-95

src/servers/tsfm/reasoning/param_space.py:25

Est = getattr(importlib.import_module(module), cls)

MEDIUMCode injection · code.dynamic_import · CWE-78, CWE-94, CWE-95

src/servers/tsfm/substrate/resolver.py:32

return getattr(importlib.import_module(module), name)