MCP server for interacting with Terra.Bio workspaces via FISS API. Provides tools for listing workspaces, querying data tables, and monitoring workflow submissions.
Terra.Bio MCP server has generally well-structured tools with mostly complete schemas and descriptions, but significant gaps in parameter documentation, error handling guidance, and output schema clarity limit its score. All 10 tools are explicitly registered with names, descriptions, and input schemas. Tool naming follows verb_noun patterns (list_*, get_*, query_*, abort_*) appropriately. However, 8 of 10 tools lack explicit output schema documentation, parameter descriptions are often minimal (e.g., 'Workspace namespace' without format guidance), and error handling provides no recovery guidance. The server implements no tool annotations (readOnlyHint/destructiveHint despite risk levels being declared), no pagination for list tools, and the one write-capable tool (abort_submission) has no confirmation pattern. Descriptions average ~140 chars, which is adequate but below production baseline of 194 chars.
Abort a running workflow submission in a Terra workspace. This will stop all currently-running workflows in the submission. Requires write access enabled on the server.
Extract specific data from Cromwell workflow metadata using flexible path notation. Supports dot notation, array indexing, and wildcards to query nested metadata structures. Useful for accessing specific task outputs, runtime attributes, or error details.
Get detailed information about a specific entity in a Terra workspace data table
Get detailed status of a workflow submission, including individual workflow statuses and summary
Estimate workflow execution costs by querying Google Cloud Batch job information for a submission. Returns estimated costs for compute resources used.
Get detailed Cromwell metadata for a specific workflow execution, including task status, timing, and logs. Includes smart summary extraction and optional GCS log retrieval.
No output schemas documented for any tool. Tools return structured data (e.g., _build_metadata_summary), but LLMs cannot see what fields to expect. This forces trial-and-error parsing and wastes tokens extracting nested data.
No pagination support for list tools (list_workspaces, list_entities, list_submissions). Without limit and offset/cursor parameters and total_count returns, large result sets will blow context windows. Baseline pattern requires this for discovery tools.
No tool annotations despite risk levels being declared (READ_ONLY vs WRITE). The abort_submission tool should have destructiveHint=true to warn agents of irreversible effects. Missing readOnlyHint for 9 read tools.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
List entities in a data table within a Terra workspace. Returns entity names and types.
List workflow submissions in a Terra workspace
List all Terra.Bio workspaces accessible to the authenticated user
Query entities in a Terra workspace using advanced filtering syntax. Supports attribute filtering with comparison operators (==, !=, <, >, <=, >=) and boolean operators (&&, ||, !). Returns matching entities with specified attributes.
No error guidance or recovery instructions. abort_submission will fail if ALLOW_WRITES=False, but the error message is generic and does not guide the agent to enable the flag or contact an admin. Lists error types but provides no actionable next steps.
Parameter descriptions lack specificity. E.g., 'Workspace namespace (e.g., broad-institute)' includes an example that LLMs may reuse literally. Descriptions do not specify format, constraints, or valid value patterns needed for intelligent input validation.
list_submissions has optional 'limit' param but no offset/cursor. If limit=10 is default and workspace has 500 submissions, LLMs cannot paginate beyond the first 10. Implement offset+limit or cursor-based pagination.
No confirmation or dry-run pattern for abort_submission. Aborting a multi-hour workflow run is destructive and non-reversible. LLMs should be able to preview impact before committing, or confirmation should be explicit.
query_entities filter_expression requires free-form string syntax. LLMs often malform filter expressions (wrong operators, missing quotes). Should provide enum-based filtering or structured filter objects to reduce parse errors.
get_workflow_metadata and extract_workflow_data return complex nested Cromwell metadata. No guidance on which fields are most relevant for agents. Applying _build_metadata_summary internally is good, but output schema should document the summary structure for LLM planning.
Tools return references (submission_id, workflow_id, entity_name) but downstream tool parameters are not documented to accept these. If get_submission_status returns submission_id='sub-123', it's unclear if abort_submission will accept the same value without a retry lookup.