Tools

The Catio MCP exposes the tools below, grouped by what they are for. Required parameters are marked.

For what the connection is for, see Why the Catio MCP. For what the MCP cannot do, see Limitations. For prompts that use these tools well, see Prompting.

Orientation

Call these first in any workflow, to establish what workspace you are in and what it contains.

• get_workspace_info: returns the current workspace and tenant: name, description, tags, and tenant ID. No parameters.
• get_workspace_context: returns a pre-rendered summary of the workspace's context, covering requirements, company profile, and architecture context. No parameters. Call it at the start of any task involving architecture decisions or compliance checks.
• get_inventory_summary: returns component counts by category and type, relationship types, last update time, and top tag keys. No parameters.

Architecture inventory

• search_inventory_components: search for components such as services, databases, servers, and load balancers. Filters: search (substring match), category, type, product (AWS, Azure, GCP, k8s), tags, page_size, page_token.
• get_inventory_component: full detail on one component, including cloud-specific properties such as instance type, region, and ARN. Requires component_id.
• search_inventory_relationships: search dependencies, data flows, and topology between components. Filters: component_id (either direction), source_component_id, target_component_id, relationship_type, relationship_subtype, search, page_size, page_token.

Context search and citations

• search_context: hybrid semantic and keyword search over indexed workspace context, including policies, uploaded documents, questionnaire answers, and architecture notes. Returns ranked passages with stable content references for citation. Requires query. Optional top_k, default 10 and maximum 50.
• resolve_citations: resolve content references returned by search_context into full citation detail: source document, title, section path, display text, and content hash. Requires content_refs.

Archie

• ask_archie: ask Archie about architecture, costs, requirements, or blueprints, against the full workspace. Requires message. Optional chat_id and conversation_history for multi-turn conversations, and documents for grounding. Returns a processing token.
• get_archie_response: poll for the result of ask_archie or edit_blueprint_with_archie. Requires processing_token. Poll every one to two seconds until the conversation completes or fails.
• edit_blueprint_with_archie: ask Archie to refine an existing blueprint in natural language. Requires blueprint_id and edit_instructions. Returns a processing token; the edited draft comes back through get_archie_response and must then be saved with update_blueprint.

Document upload

Two steps, because file bytes do not travel through the MCP request body.

• create_upload_url: request a presigned upload URL for a PRD or reference document. Requires filename, mime_type (PDF, plain text, or Markdown), and size_bytes. Returns a file ID and an upload URL to PUT the raw bytes to.
• confirm_upload: finalize the upload once the bytes are in place. Requires file_id. Returns a document reference usable in ask_archie, create_blueprint_outline, or generate_blueprint_list.

Blueprint outlines

The reviewable path: see a lightweight outline before committing to full generation.

• create_blueprint_outline: generate a reviewable outline, being a title, description, and plan items, from a goal. Requires goal. Optional documents and view_id. Does not run full generation.
• refine_blueprint_outline: revise an outline with a guidance prompt, producing a new revision. Requires outline_id, base_revision, and refinement_prompt.
• accept_blueprint_outline: accept a revision and start generation. Requires outline_id and revision. Returns a batch ID.
• decline_blueprint_outline: decline an outline, so no generation starts. Requires outline_id.

Blueprint generation

The fast path, which skips outline review.

• generate_blueprint_list: trigger generation directly from a goal. Requires goal. Optional documents and view_id. Returns a batch ID immediately and continues in the background.
• list_recommendation_lists: list generation batches for the workspace. Optional status (SEARCHING, WRITING, COMPLETED, FAILED), page_size, page_token.
• get_recommendation_list_status: check one batch. Requires list_id. Poll until COMPLETED or FAILED.

Blueprints

Blueprints come in two subtypes: recommendation, an opportunity to improve, and design, a specification for something to build or change. See Blueprints.

• create_recommendation: create a recommendation. Requires title and category. Optional content, external_tracking_link, metadata.
• create_design: create a design. Same shape as create_recommendation.
• update_blueprint: update a recommendation or design. Read-modify-write, so only the sections you provide change. Requires blueprint_id. Optional title, category, content, external_tracking_link, metadata, is_bookmarked, is_complete.
• get_recommendation: fetch one blueprint by ID, with all its sections. Requires recommendation_id.
• search_recommendations: search blueprints with free text and structured filters. Filters: query, status, subtype, category, domain, benefit, sort_by, sort_order, page_size, page_token.

Blueprint status values

NEW, PINNED, MET, DISMISSED, SUPERSEDED

Category values

Used by create_recommendation, create_design, and update_blueprint, and available as a filter on search_recommendations. Category is the subject area of a blueprint, which is a different dimension from the benefit it delivers.

ai_platform, api_integration, application_architecture, compliance_audit, cost_optimization, data_architecture, infrastructure, infrastructure_monitoring, messaging_eventing, network_security, operational_excellence, operational_resilience, organisational_productivity, performance_optimization, product_capability, resilience_assessment, security_and_compliance, security_posture