Best for
- Maintenance engineers
- Reliability specialists
- Facility planners
- Industry 4.0 researchers
A
AssetOpsBench
Community
A multi-agent MCP benchmark and framework for industrial asset operations
GitHub source repository ↗
45FMRS · D
IBM/AssetOpsBench is an Apache-2.0 industrial asset-operations AI-agent benchmark and framework with IoT, FMSR, TSFM, work-order, and vibration tools plus multiple agent-orchestration approaches. The material confirms MCP support but does not provide a server manifest, transport protocol, authentication method, or client configuration.
AssetOpsBench is an open framework for building, orchestrating, and evaluating domain-specific AI agents for Industry 4.0 asset operations and maintenance. It provides industrial scenarios, domain MCP servers, agent frameworks, and evaluation workflows covering IoT, failure modes and sensor relationships, time-series foundation models, work orders, and vibration analysis.
Tools
sites
Lists sites.
asset_ids
Lists asset identifiers.
asset_detail
Gets asset details.
assets
Lists assets.
find_assets_by_sensors
Finds assets by sensors.
installed_sensors
Lists installed sensors.
measured_sensors
Lists sensors with measured data.
get_failure_modes
Gets failure modes.
generate_failure_modes
Generates failure modes.
add_failure_modes
Adds failure modes.
generate_failure_mode_sensor_mapping
Generates failure-mode-to-sensor mappings.
list_tasks
Lists time-series tasks.
profile_series
Profiles time series.
characterize_series
Characterizes time series.
data_quality
Checks data quality.
list_models
Lists models.
search_models
Searches models.
find_models
Finds models.
resolve_model
Resolves a model.
model_template
Provides a model template.
register_model
Registers a model.
register_finetuned
Registers a fine-tuned model.
hf_stats
Gets Hugging Face statistics.
list_features
Lists features.
search_features
Searches features.
extract_features
Extracts features.
select_features
Selects features.
recipe_template
Provides a recipe template.
run_recipe
Runs a recipe.
run_tabular_recipe
Runs a tabular recipe.
run_plan
Runs a plan.
evaluate
Evaluates run results.
list_runs
Lists runs.
list_results
Lists results.
get_work_order_distribution
Gets work-order distributions.
predict_next_work_order
Predicts the next work order.
compute_fft_spectrum
Computes an FFT spectrum.
compute_envelope_spectrum
Computes an envelope spectrum.
Setup
From the repository root, run: git clone https://github.com/IBM/AssetOpsBench.git; cd AssetOpsBench; pip install -e .。MCP servers and the plan-execute runner are located under src/; full setup, MCP server, and plan-execute instructions are provided in INSTRUCTIONS.md. The README marks the example scenario as to be enabled.
Fit and risk
Not for
- Users requiring verified production connectors, authentication, or transport details.
- Users seeking general-purpose chat or non-industrial asset-maintenance tools.
- Users expecting to run the complete scenarios without consulting the project setup documentation.
Required permissions
- Requires access to the simulated industrial environment data used by the project.
- Depending on the task, it may need to read sensor, asset, FMEA, alert, and work-order records.
- Specific authentication, authorization, and write permissions are not described in the provided material.
Risks and side effects
- The server manifest is unavailable, so the actual tool surface, startup method, transport, and authentication details require confirmation.
- The project is under active development; the main branch may differ from branches used for publications.
- Operations such as adding failure modes, registering models, and running recipes may change records in the project environment; permissions and rollback behavior are not described.
Troubleshooting
- Confirm that IBM/AssetOpsBench is cloned and run pip install -e . from the repository root.
- Check that the Python environment and dependencies are installed.
- Read INSTRUCTIONS.md for MCP server and plan-execute runner configuration.
- Verify that the selected branch contains the required code; the README identifies main as the active development branch and names separate publication branches.
- If the example scenario fails, note that the README labels it as to be enabled.
Use cases
Query industrial sites, assets, and sensor information.
Analyze failure modes, sensor relationships, and time-series data.
Find, run, and evaluate time-series models and feature recipes.
Explore multi-agent orchestration and evaluation for industrial asset maintenance.
Supported clients
Supported clients have not been confirmed yet.