OpenConstruction MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| HF_TOKEN | No | Hugging Face token for accessing private datasets. | |
| SUPABASE_URL | No | Supabase URL. Required when running the remote MCP/API server. | |
| OC_PUBLIC_URL | No | Public URL of the server, used for OAuth callback URLs. | |
| OC_DOWNLOAD_ROOT | No | Root directory for dataset downloads. Defaults to ~/.openconstruction/datasets. | ~/.openconstruction/datasets |
| SUPABASE_ANON_KEY | No | Supabase anonymous key. Required when running the remote MCP/API server. | |
| OC_MAX_DOWNLOAD_BYTES | No | Maximum total streamed HTTP transfer size per download. Defaults to 500 GiB. | 500 GiB |
| OC_TOKEN_ENCRYPTION_KEY | No | Encryption key for provider tokens. Required when running the remote MCP/API server and for connected accounts. | |
| OC_PAPER_CONTENT_BASE_URL | No | Base URL for paper PDF downloads. Defaults to GitHub media.githubusercontent.com for the default paper registry. | https://media.githubusercontent.com/media/Ben11304/OC-clone/open-construction-data/papers |
| OC_CONNECTED_ACCOUNTS_ENABLED | No | Set to 'true' to enable Connected Accounts. Defaults to 'false'. | false |
| OPENCONSTRUCTION_DATA_BASE_URL | No | Override the catalog endpoint. Defaults to the public catalog snapshot at https://github.com/Ben11304/OC-clone/tree/main/open-construction-data. | https://github.com/Ben11304/OC-clone/tree/main/open-construction-data |
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_resourcesC | Search datasets, models, workflows, OERs, tools, guides, contributors, benchmarks, and vocabulary. |
| get_resourceB | Get one resource by type and id. |
| compare_resourcesC | Compare two or more catalog resources for a user goal. |
| get_catalog_statsA | Return counts by resource type for the loaded catalog snapshot. |
| ask_openconstructionB | Answer a catalog-grounded question with matching resources. |
| find_datasetsC | Find datasets by task, modality, object class, annotation type, license, or access needs. |
| run_dataset_discoveryC | Run the Dataset Discovery skill as a structured workflow that searches, ranks, and explains dataset candidates. |
| find_modelsC | Find models by task, dataset, modality, method, or publication evidence. |
| explain_schemaC | Explain common OpenConstruction metadata fields and where they appear in catalog records. |
| analyze_catalog_gapsB | Inspect catalog coverage by type and return possible metadata gaps for a topic. |
| prepare_benchmark_submissionD | Create a checklist for preparing benchmark metadata and linked resources. |
| validate_metadata_recordB | Check one draft metadata record for important missing fields. |
| list_skillsC | List repo-defined OpenConstruction skill metadata. |
| get_skillB | Get one OpenConstruction skill definition by id. |
| get_dataset_download_planD | Resolve the trusted OpenConstruction metadata route without writing files, including local provider-auth guidance when required. |
| download_datasetA | Start a background download inside OC_DOWNLOAD_ROOT. If status is auth_required, show its local provider login instructions and retry payload; never ask the user to paste credentials into chat. Requires explicit license acceptance and never accepts arbitrary source URLs or shell commands. |
| get_download_statusA | Return progress, files, and errors for a local OC dataset download job. |
| cancel_downloadB | Request cancellation of a running local OC dataset download job. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 18 tools
Several tools overlap in the discovery/search space: search_resources, find_datasets, find_models, and run_dataset_discovery all perform similar resource finding/ranking workflows. Meanwhile, ask_openconstruction serves a distinct Q&A purpose and compare_resources is distinct. Tool descriptions help disambiguate, but the boundaries between search and discovery still require careful reading.
All 18 tools follow the predictable snake_case verb_noun pattern (list_skills, get_skill, download_dataset, get_download_status, search_resources, etc.), with stylistic exceptions like run_dataset_discovery and get_catalog_stats, but still consistent and readable.
18 tools is on the higher end but acceptable for a broad OpenConstruction MCP server covering datasets, models, workflows, resources, benchmarks, toolkit, and vocabulary. The domain is broad and each tool gives a useful and non-superfluous facet, though consolidation (e.g., combining search-related tools) would make it more efficient.
The surface covers end-to-end datasets/resource discovery: search, list, detail, download, status and cancellation, plus workflows, comparison, gap analysis, metadata validation, benchmarking, and vocabulary explanation. Minor practical gaps exist (e.g., no generic agent update/delete endpoint, no archive metadata Refresh), but the core contribution path is present.