MCP tool reference

The list below is generated from @repo/worker-runtime's MCP tool catalog, the same data document-server.ts passes to registerTool when it starts the MCP server for a turn. Every name, parameter, and availability note here matches what your agent actually receives — there is no separate hand-written tool description to fall out of sync.

Expand a tool to see its parameters. Required parameters must be present in the call; optional parameters can be omitted.

Availability

Most tools are registered on every document MCP server the runtime starts. A few are conditional:

get_task_details and list_task_comments require the server to start with a client, current taskId, and workspaceId. The orchestrator always supplies these for a workflow-linked task turn, so both tools are present on every real turn, even when the task has no linked documents. get_task_details takes no input — it always resolves to the current task, so an agent cannot point it at another task by supplying a different ID.
get_parent_details requires the same client/taskId/workspaceId context plus a parent task linked to the current task via a part_of relation. It also takes no input and is absent from tasks with no parent.
list_linked_documents and read_linked_document are registered only when the task has at least one manually-linked document (a spec, requirement, or similar — not a workflow artifact like plan/research/review/code_summary, which the document-block tools below already cover). Absent on tasks with no linked documents.
report_workflow_verdict is registered only for review-mode workflow task turns. It is absent from plan and code turns.
report_result is registered whenever the host supplies a result handler. The orchestrator supplies one for every task mode, so it is present on every real turn.
Prompt every successful mode to call report_result once with a one- or two-sentence, non-technical owner summary. Only Execute requires the call; other current and future modes use the normal final message when it is omitted or fails. Review still reports its structured verdict separately.

See documents, tools, and artifacts for how the document-block tools, review verdicts, and result summaries fit into a turn, and the ACP contract for how session/new delivers this MCP server to your agent.

Documents & tools

The MCP tools every runtime receives: workflow artifacts, review verdicts, result summaries, and visual artifacts.

Runtime manifest

The one.json reference: everything the worker runtime needs to install, verify, and launch your agent.

ACP contract

How the worker runtime drives your agent over the Agent Client Protocol: turn lifecycle, headless behaviors, trust boundaries, and outcome mapping.

Reference

Reference runtime responsibilities, workflow context, artifact documents, environment variables, and API operation families.

MCP tool reference

Expand a tool to see its parameters. Required parameters must be present in the call; optional parameters can be omitted.

Availability

Most tools are registered on every document MCP server the runtime starts. A few are conditional:

get_task_details and list_task_comments require the server to start with a client, current taskId, and workspaceId. The orchestrator always supplies these for a workflow-linked task turn, so both tools are present on every real turn, even when the task has no linked documents. get_task_details takes no input — it always resolves to the current task, so an agent cannot point it at another task by supplying a different ID.
get_parent_details requires the same client/taskId/workspaceId context plus a parent task linked to the current task via a part_of relation. It also takes no input and is absent from tasks with no parent.
list_linked_documents and read_linked_document are registered only when the task has at least one manually-linked document (a spec, requirement, or similar — not a workflow artifact like plan/research/review/code_summary, which the document-block tools below already cover). Absent on tasks with no linked documents.
report_workflow_verdict is registered only for review-mode workflow task turns. It is absent from plan and code turns.
report_result is registered whenever the host supplies a result handler. The orchestrator supplies one for every task mode, so it is present on every real turn.
Prompt every successful mode to call report_result once with a one- or two-sentence, non-technical owner summary. Only Execute requires the call; other current and future modes use the normal final message when it is omitted or fails. Review still reports its structured verdict separately.

Documents & tools

The MCP tools every runtime receives: workflow artifacts, review verdicts, result summaries, and visual artifacts.

Runtime manifest

The one.json reference: everything the worker runtime needs to install, verify, and launch your agent.

ACP contract

How the worker runtime drives your agent over the Agent Client Protocol: turn lifecycle, headless behaviors, trust boundaries, and outcome mapping.

Reference

Reference runtime responsibilities, workflow context, artifact documents, environment variables, and API operation families.

Availability

select_design_doc_diagramsConditionalPick useful plan diagrams (including Wiremark UI). Returns a compact JSON list; empty means prefer prose or a table.

get_design_doc_diagram_detailsConditionalReturn guidance and templates for diagram IDs (includes offline Wiremark rules for wiremark_ui).

validate_design_doc_diagramConditionalValidate Wiremark or Mermaid source before embedding. Pass fence body only (no ```). Returns valid, line errors, and warnings.

lookup_wiremark_componentConditionalLook up a Wiremark component by name or query. Returns Markdown.

get_research_guidanceConditionalReturn deterministic outputs, questions, criteria, and conditional sections for selected research methods.

get_research_graph_guidanceConditionalReturn candidate diagram identifiers for selected research methods. Evidence decides whether any candidate is used.

get_review_guidanceConditionalReturn deterministic target focus, evidence relationships, methodology, questions, and criteria for a Review step.

list_artifact_blocksAlways availableList block metadata from declared workflow artifacts without loading bodies.

load_artifact_blocksAlways availableLoad artifact blocks with their markdown bodies.

set_artifact_titleAlways availableSet the document title for a declared writable workflow artifact. Call before or with the first block write.

append_artifact_blockAlways availableAppend a block to a declared writable artifact. Creates the artifact if missing.

replace_artifact_block_bodyAlways availableReplace an existing artifact block's markdown body.

update_artifact_block_metadataAlways availablePatch YAML metadata on an existing artifact block (status, severity, title, …).

find_related_blocksAlways availableFollow related refs across declared artifacts (qualified refs: documentId:blockId).

get_task_detailsConditionalFetch this turn's task with full description. No input; always the current task. Use list_task_comments / read_linked_document for comments and linked docs.

get_parent_detailsConditionalFetch this turn's parent initiative with full description. No input; always the linked parent.

list_task_commentsConditionalList comments on this turn's task. Excludes agent_runtime recaps unless includeAgentRecaps is true.

list_linked_documentsConditionalList manually linked docs (specs, requirements, …). Not workflow artifacts — use list_artifact_blocks for those. Pass documentId to read_linked_document.

read_linked_documentConditionalRead a linked document by documentId from list_linked_documents. Rejects IDs not linked to this task.

request_inputConditionalPause the workflow turn until a human responds. Call when you cannot continue without human judgment, credentials, or a product decision. Do not call for fixable issues the next agent can handle.

report_workflow_verdictReview turns onlySubmit the review-step verdict (do not write a text verdict). summary posts as the task comment; artifacts select blocks for the next step prompt.

report_resultConditionalSave a one- or two-sentence, non-technical owner summary of what was done or concluded and the outcome. Call once when work is done.

Documents & tools

Runtime manifest

ACP contract

Reference

Availability

select_design_doc_diagramsConditionalPick useful plan diagrams (including Wiremark UI). Returns a compact JSON list; empty means prefer prose or a table.

get_design_doc_diagram_detailsConditionalReturn guidance and templates for diagram IDs (includes offline Wiremark rules for wiremark_ui).

validate_design_doc_diagramConditionalValidate Wiremark or Mermaid source before embedding. Pass fence body only (no ```). Returns valid, line errors, and warnings.

lookup_wiremark_componentConditionalLook up a Wiremark component by name or query. Returns Markdown.

get_research_guidanceConditionalReturn deterministic outputs, questions, criteria, and conditional sections for selected research methods.

get_research_graph_guidanceConditionalReturn candidate diagram identifiers for selected research methods. Evidence decides whether any candidate is used.

get_review_guidanceConditionalReturn deterministic target focus, evidence relationships, methodology, questions, and criteria for a Review step.

list_artifact_blocksAlways availableList block metadata from declared workflow artifacts without loading bodies.

load_artifact_blocksAlways availableLoad artifact blocks with their markdown bodies.

set_artifact_titleAlways availableSet the document title for a declared writable workflow artifact. Call before or with the first block write.

append_artifact_blockAlways availableAppend a block to a declared writable artifact. Creates the artifact if missing.

replace_artifact_block_bodyAlways availableReplace an existing artifact block's markdown body.

update_artifact_block_metadataAlways availablePatch YAML metadata on an existing artifact block (status, severity, title, …).

find_related_blocksAlways availableFollow related refs across declared artifacts (qualified refs: documentId:blockId).

get_task_detailsConditionalFetch this turn's task with full description. No input; always the current task. Use list_task_comments / read_linked_document for comments and linked docs.

get_parent_detailsConditionalFetch this turn's parent initiative with full description. No input; always the linked parent.

list_task_commentsConditionalList comments on this turn's task. Excludes agent_runtime recaps unless includeAgentRecaps is true.

list_linked_documentsConditionalList manually linked docs (specs, requirements, …). Not workflow artifacts — use list_artifact_blocks for those. Pass documentId to read_linked_document.

read_linked_documentConditionalRead a linked document by documentId from list_linked_documents. Rejects IDs not linked to this task.

request_inputConditionalPause the workflow turn until a human responds. Call when you cannot continue without human judgment, credentials, or a product decision. Do not call for fixable issues the next agent can handle.

report_workflow_verdictReview turns onlySubmit the review-step verdict (do not write a text verdict). summary posts as the task comment; artifacts select blocks for the next step prompt.

report_resultConditionalSave a one- or two-sentence, non-technical owner summary of what was done or concluded and the outcome. Call once when work is done.

Documents & tools

Runtime manifest

ACP contract

Reference