create_device
ChatGPTCreate a device and its components in the selected organization. Return the created device with its IDs.
create_document
ChatGPTCreate a new document shell. Provide exactly one of template_key or default_title. Template-backed documents use template_key; freeform documents use default_title. After this returns, use create_document_version if you need an editable draft and edit_document to change document content.
create_document_version
ChatGPTCreate or reuse an editable draft for a Dovetail document. Use this after read_document_metadata shows the latest version is a finalized revision and you need an editable draft before edit_document or edit_document_comments. If an editable latest draft already exists, this tool returns it without creating another version. This tool rejects uploaded documents, documents that do not support live editing, and versions currently routing for approval.
create_quality_event_record
ChatGPTCreate a workflow record immediately after the user authorizes its content. Read the definition first. Creation starts at the first step; incomplete answers and an empty title are allowed. Later steps start empty. Supply a stable client-generated record.id UUID and reuse it when reconciling an uncertain result. Never request reviewers during creation or make approval decisions. Get responsible user IDs from get_team_members (follow nextSkip). This is a direct write, without an in-app draft/Save form. Do not invent facts or deadlines. On a conflict or timeout, read the same record before considering another write; never generate a replacement ID to retry.
create_skill
ChatGPTCreate a private skill with markdown instructions for your own use. The name must be lowercase kebab-case. Confirm the write by setting confirmed to true. New skills are always private and enabled by default for their creator. System wide force enabling is unavailable.
create_traceability_matrix_link
ChatGPTCreate a traceability matrix link. Use this after reading the traceability matrix and identifying the exact source and target IDs to connect. Supported link_type values: - risk_control_ssr: provide rc_id and ssr_id - ssr_design_file: provide ssr_id and document_id - test_validation_file: provide test_id and document_id
delete_quality_event_record
ChatGPTDelete an accidentally created CAPA after the user authorizes deletion. Read the record first and identify its title and ID to the user. Only an editor can delete it, and only before Evaluate has ever been approved. The backend enforces these rules and removes linked workflow tasks. If the outcome is uncertain, read the same record before reporting its state; do not retry deletion automatically.
delete_traceability_matrix_link
ChatGPTDelete a traceability matrix link. Use this after reading the traceability matrix and identifying the exact source and target IDs to disconnect. Supported link_type values: - risk_control_ssr: provide rc_id and ssr_id - ssr_design_file: provide ssr_id and document_id - test_validation_file: provide test_id and document_id
edit_data_key
ChatGPTUpdate the value of a data key. The tool resolves the current source-answer context, validates the proposed final value, and asks the backend to write it. The backend is authoritative for whether the selected version can be modified. When accepted, the new value takes effect immediately and overwrites the current one. Use exactly one mode: - Full replacement: provide value - Partial edit: provide old_string and new_string, optionally with replace_all=true Rules: - choose one explicit source object for either a document version or staged assistant-process state - an assistant-process source contains only process and access scope; document/version fields belong only to a document source - the selected version must be an editable draft; the backend rejects finalized versions and versions routing for approval - for a finalized version, call create_document_version first; for a routing version, resume editing before changing the data key - partial edits are case-sensitive - partial edits fail if old_string is missing, not found, empty against a non-empty current value, or matches multiple times unless replace_all=true Returns the identifier of the document version that was updated.
edit_document
ChatGPTPass the Dovetail document UUID as document_id. Edit the latest live draft as tracked changes. Historic revisions cannot be edited. Do not provide meta justifications. Edit the document. <tool-input> operations: an array of edit operations. Each operation is an object with type, target, and content properties. <type> replace (replaces the target), insertBefore (inserts content before the target), insertAfter (inserts content after the target) </type> <target> The target is the content where the operation will be applied. It can be: - a 6-character hash identifier of a node (obtained from the _hash attribute in nodes returned by read_document tool). Always use the value of the _hash attribute itself, never the node's id or any other attribute. Each operation can only target one node at once. To edit multiple nodes, create one operation for each node. - "doc" to target the entire document. Use it to replace the entire document with new content. </target> <content> Should be a valid Dovetail document JSON fragment (an array of Dovetail document JSON nodes), not a single node object. Do not include _hash attributes. </content> To delete a node, replace it with an empty array. </tool-input> <tool-result> operationResults: the results of applying the operations </tool-result> <important-rules> Call the read_document tool to get the _hash attributes of the nodes you want to edit. In the insertBefore and insertAfter operations, the content should not contain the current content of the node, only the new content that is inserted before or after. </important-rules>
edit_skill
ChatGPTEdit a skill and optionally enable or disable it for your account. Supply at least one field and set confirmed to true. Name, markdown, and visibility edits require creator permission; enabling or disabling uses your own preference. Backend permissions determine which skills and visibilities you may edit. System wide force enabling is unavailable.
export_quality_event_record
ChatGPTExport a saved quality event record as a PDF or DOCX file. Use the organization and record IDs from the saved record. The returned embedded resource contains the file; no document data is changed.
get_510k
ChatGPTGet details for a specific 510(k) number from the FDA database. Use this to inspect the device record and retrieve the parsed summary URL that can then be passed to read_summary.
get_agent_run_result
ChatGPTCheck the status of a previously started AGENT run. Pass the agent_run_id returned by one of the run_*_agent tools. The result is either working, completed with final output, or failed with an error message.
get_assistant_process
ChatGPTGet the complete database entry for an assistant process. Use this when the user refers to a staged or proposed value. The process state contains changes that have not been applied to documents yet, so inspect it before searching document metadata for that value. Provide the process, organization, and device IDs that identify the process context.
get_dashboard
ChatGPTGet the computed dashboard for an organization or device. Omit device_id for the organization dashboard. Provide device_id when the user wants the dashboard scoped to a specific device.
get_device
ChatGPTGet basic device information and characteristics. Returns high-level device metadata such as name, description, market, and components. Prefer read_data_key or read_document if you need specific regulatory content rather than an overview.
get_devices
ChatGPTGet the devices the authenticated user can access in one organization. Use this to identify the device ID you need before reading data keys, documents, templates, or running an AGENT against a specific device. Call get_organizations first if you need an organization ID.
get_functional_groups
ChatGPTGet the functional groups for an organization. Use this when you need the available functional groups before creating or updating documents that belong to a specific function or team.
get_organizations
ChatGPTGet the organizations the authenticated user can access. Use this to understand which organizations are available to the current user before listing or working with devices and related resources.
get_tasks
ChatGPTGet tasks for the authenticated user or the whole organization. scope="mine" returns tasks assigned to the current user. Use scope="all" when the user explicitly needs organization-wide task visibility. Provide device_id to include tasks for that device plus organization-wide tasks.
get_team_members
ChatGPTGet team members for an organization. Use this when you need collaborators in a specific organization, for example to identify reviewers or document owners across that org.
get_traceability_matrix
ChatGPTGet the device traceability matrix payload. Returns the backend traceability matrix JSON for a device, including chains, relations, and Mermaid definitions. Set plain_mermaid=true to remove Mermaid styling from chains[].mermaidDefinition. Mermaid click links are omitted by default because MCP clients cannot use them; set include_mermaid_links=true to include source links.
grep_search_documents
ChatGPTRun a fast regex search across document and data key content. This is best for finding exact text matches or regex patterns and is more precise than manually scanning with read_document or read_data_key. The query must be a valid regex, so escape special characters when needed.
grep_search_standards
ChatGPTRun a fast regex search across standards content. This is best for finding exact text matches or regex patterns inside standards. The query must be a valid regex, so escape special characters when needed.
list_data_key_changes
ChatGPTList audited data-key value changes for a device, grouped by document. Returns the facts recorded for every data-key answer INSERT, UPDATE or DELETE on the device's documents in the half-open window (since, until]: since is exclusive, until is inclusive, so consecutive calls that pass the previous call's returned until as the next since neither overlap nor gap. Per change: the data key, the document and version it was written in, the old and new values, when it happened, who made the change, and which other documents consume the value (live drafts that currently resolve to this answer, and frozen snapshots that baked it). Answers copied into a new draft from a finalized revision are not listed; only real edits are. since defaults to 24 hours ago; pass an explicit window for anything scheduled. until defaults to the request time, resolved once per call; the resolved window is returned as since/until. An explicit bound must carry a zone (Z or an offset): a value without one is rejected, as the backend route and the agents tool reject it, instead of being read as UTC. The editor of each change is labelled client, dovetail_staff (formly.ai / withdovetail.com email) or unknown (no human identity recorded). This tool returns ALL authorized history by default: nothing is hidden, and excluded_dovetail_staff_changes is 0. Only when exclude_dovetail_staff=true (which the change digest passes) are dovetail_staff edits dropped and counted; unknown editors are always included. When the backend could not read a candidate consumer draft, warnings lists it and the affected changes carry consumers_incomplete: true with a "consumers may be incomplete: <titles>" note. A window holding more than 5000 changes is rejected; narrow since/until and call again. Values longer than 50,000 characters are truncated with a summary, like read_data_key. Read-only: this tool makes no recommendations and changes nothing.
list_data_keys
ChatGPTList available data-key definitions. A data-key id can belong to multiple document records. The id identifies the field definition; each document version holds its own value. Use a selected document_id/version_id pair when reading or editing one.
list_document_templates
ChatGPTList document templates available for the configured device. Use this if you need to understand which template keys are available in the system before reading a template or creating related content.
list_documents
ChatGPTList all available documents for a device in JSON format. Returns a lightweight list with title, id, and templateKey. Use read_document_metadata for revision metadata and read_document for full document content.
list_quality_event_definitions
ChatGPTList available workflow quality events, such as CAPA, in an organization. Select org_id from get_organizations. No device is required.
list_quality_event_document_references
ChatGPTList selectable identities for quality event document_reference answers. Each item contains documentId, documentVersionId and displaySnapshot. Use the returned object unchanged rather than inventing document IDs.
list_quality_event_files
ChatGPTList existing organization evidence files for quality event answers. File answers use arrays of returned file UUIDs. This does not upload files.
list_quality_event_records
ChatGPTFind saved workflow records in an organization, optionally by quality event, status or title. Returns saved answers, definitions, versions and workflow state. Select org_id from get_organizations; record IDs come from these results.
list_skills
ChatGPTList skills visible to you in an organization. Results include skill IDs, descriptions, visibility, and your enabled state, including skills you disabled. Use mine to find skills you created, then check canEdit before editing. Search matches names and descriptions. Markdown bodies are not included in this list; use read_skill for the full content.
list_standards
ChatGPTList all standards available to the current server.
memory_recall
ChatGPTSearch recalled expert knowledge relevant to the current device. This looks across organization-specific memory and the configured expert knowledge namespace to return concise retrieved context for the query.
read_data_key
ChatGPTGet the value for a given data key. By default this returns the first 50,000 characters with a summary when the value is longer. Choose a document source with document_id and version_id, or an assistant-process source with assistant_process_id, org_id, and device_id.
read_data_key_configuration
ChatGPTGet the configuration for a specific data key. Use this to inspect the data key type, guideline text, related data keys, and visibility rules before reading or editing the value.
read_document
ChatGPTRead the full body for a document version. Live editable drafts return the current Dovetail document JSON. Stored revisions return the saved Dovetail document JSON when available. Uploaded documents return their stored parsed markdown. If version_id is omitted, the latest draft is preferred and otherwise the latest retrievable revision is returned. For Dovetail document JSON, from_node starts reading at a zero-based top-level node index. Use read_document_metadata for version, author, and approver metadata.
read_document_metadata
ChatGPTRead document metadata for a document and its retrievable versions. Use this to inspect retrievable versions, authors, approvers, and revision details without fetching the full document body.
read_document_template
ChatGPTLook up a document template and the data keys that are part of it. Use this to understand which template keys and fields are part of a given document so you can create or review the relevant content sequentially. Always call read_data_key_configuration separately for each returned data key before filling out answers; the template response only identifies the fields, while the data key configuration provides the field-specific guidance, type, context, and constraints needed to complete them correctly.
read_quality_event_definition
ChatGPTRead a quality event's ordered steps, fields, requirements and approval definitions. Read this before creating a record. Use its first step for creation; never invent field keys, evidence, identities, deadlines or decisions.
read_quality_event_record
ChatGPTRead a record's current version, definition, answers and approval state. Read before editing or reconciling an uncertain write. Preserve row IDs and unrelated answers. Only backend results establish completion or closure. The record owner is createdBy. A step's owner=null means that creator is responsible, not that the step is unassigned. Resolve alternate owners and approvers with get_team_members, following nextSkip.
read_quality_event_record_activity
ChatGPTRead saved record activity, including changed answers and review decisions.
read_skill
ChatGPTRead a visible skill's full markdown instructions and metadata. Use list_skills to find the skill ID. A skill you disabled remains readable if you can still access it. The backend checks organization membership and skill visibility.
read_standard
ChatGPTRead a standards file by ID. Returns a paged view of the standard content. Use this to review the actual text of a standard after locating the file via list_standards or grep_search_standards.
read_summary
ChatGPTRead the text content of a parsed 510(k) summary file. Pass the Parsed PDF URL returned by search_510k_summaries or get_510k. Do not use the original FDA 510(k) URL here.
resume_agent_run
ChatGPTContinue a completed agent conversation with an answer or follow-up. A question in the final reply is a normal completed run. Send your answer here to start a new run in the same saved conversation and receive a new handle to poll with get_agent_run_result. Use the original handle unchanged. Conversations have a 24-hour retention period. This tool does not approve pending actions or resume interrupted runs.
run_expert_reviewer_agent
ChatGPTStart the Expert Reviewer and return an agent_run_id. Use this when you need an agentic expert review of regulatory content or reasoning for a specific device. Specify the device name or ID and the review goal in request. For an organization-wide review, state that scope explicitly. The agent can discover the device or ask for clarification if needed. This call starts the background run only. Use get_agent_run_result with the returned handle until the status is completed.
run_predicate_researcher_agent
ChatGPTStart the Predicate Researcher and return an agent_run_id. Use this when you need agentic predicate or FDA research for a specific device. Specify the device name or ID and the research goal in request. The agent can discover the device or ask for clarification if needed. This call starts the background run only. Use get_agent_run_result with the returned handle until the status is completed.
run_standards_and_guidances_researcher_agent
ChatGPTStart the Standards and Guidances Researcher and return an agent_run_id. Use this when you need agentic research across standards and guidances for a specific device. Specify the device name or ID and the research goal in request. The agent can discover the device or ask for clarification if needed. This call starts the background run only. Use get_agent_run_result with the returned handle until the status is completed.
search_510k_summaries
ChatGPTSearch 510(k) summaries by semantic similarity to the query. The query should be a short description of the device and the relevant characteristics you want to match. Results include similar devices and metadata such as organization, product code, received year, FDA URL, parsed summary URL, and similarity score.
submit_feedback
ChatGPTSend product feedback about Dovetail MCP to the Dovetail team in Slack. Include the problem, expected behavior, and relevant tool name. Do not send credentials, patient data, or document contents. This creates a message, not a support ticket or an automatic fix. Do not retry an uncertain result.
update_quality_event_record
ChatGPTSave user-authorized edits to the current or previous steps in one update. Read the record first and use its reviewed version as expected_version. Omitted fields stay unchanged; null clears nullable values. Answer arrays replace the entire array: include every retained row, saved row ID and unchanged row answer. Additional step edits go in changes.steps; all save atomically. New repeatable rows need client-generated UUIDs. Reviewer selections use {approver: user UUID, dueDate?: date or null} inside the definition's approval fields, and only when requested. Existing reviews cannot be cancelled; content edits may invalidate them. Never send approval decisions, request IDs, status or provenance. Only the selected reviewer decides in Dovetail. Future and closed steps are locked. This is a direct write, without an in-app Save form. On a stale version or uncertain outcome, read again and review changes before another write.
create_device
Claudecreate_document
Claudecreate_document_version
Claudecreate_quality_event_record
Claudecreate_skill
Claudecreate_traceability_matrix_link
Claudedelete_traceability_matrix_link
Claudeedit_data_key
Claudeedit_document
Claudeedit_skill
Claudeget_510k
Claudeget_agent_run_result
Claudeget_assistant_process
Claudeget_dashboard
Claudeget_device
Claudeget_devices
Claudeget_functional_groups
Claudeget_organizations
Claudeget_tasks
Claudeget_team_members
Claudeget_traceability_matrix
Claudegrep_search_documents
Claudegrep_search_standards
Claudelist_data_key_changes
Claudelist_data_keys
Claudelist_document_templates
Claudelist_documents
Claudelist_quality_event_definitions
Claudelist_quality_event_document_references
Claudelist_quality_event_files
Claudelist_quality_event_records
Claudelist_skills
Claudelist_standards
Claudememory_recall
Clauderead_data_key
Clauderead_data_key_configuration
Clauderead_document
Clauderead_document_metadata
Clauderead_document_template
Clauderead_quality_event_definition
Clauderead_quality_event_record
Clauderead_quality_event_record_activity
Clauderead_skill
Clauderead_standard
Clauderead_summary
Clauderesume_agent_run
Clauderun_expert_reviewer_agent
Clauderun_predicate_researcher_agent
Clauderun_standards_and_guidances_researcher_agent
Claudesearch_510k_summaries
Claudesubmit_feedback
Claudeupdate_quality_event_record
Claude