Dutify

npx add-mcp 'https://mcp.dutify.ai'

This distribution now covers Hub, PM, Wiki and Roadmarq. Existing installations retain the skill name dutify-api; Hub-only installations can keep dutify-hub-api. Personal du_live_ keys work across granted products; old workspace keys retain product boundaries. See routing and connector compatibility. The combined MCP route is adopted only after its separate release and tool-list verification.

dutify-api — Claude Code skill

Version 2026.09.25.2 adds capability-aware guidance for dw_live_ workspace credentials. Read integration keys before using that family; updating instructions does not deploy the authority or consumers. The authority makes issuance available immediately when deployed.

A Claude Code (and Claude.ai) skill that teaches an LLM how to use the Dutify HTTP API directly: discover endpoints via the aggregated catalog at https://dutify.ai/mp/api/v1/api-catalog, call the right "lite" tag with an X-API-Key, and self-correct on validOptions errors instead of guessing endpoint shapes from memory.

Covers Hub, Project Management, Wiki/Codexum, and Roadmarq in one skill. Hub uses its catalog at https://dutify.ai/api/v1/api-catalog; the existing Suite catalog aggregates PM, Wiki and Roadmarq.

Install

User-level (any Claude Code session, any project)

git clone https://github.com/dutifyai/dutify-cloud-ai-skill.git ~/.claude/skills/dutify-api

After install, restart your Claude Code session (or /clear). The skill will auto-load when a prompt mentions Dutify ("create a task in…", "list bugs assigned to…", "what wiki pages live under…", etc.) — no need to invoke it explicitly.

Project-level (only when working inside one project)

git clone https://github.com/dutifyai/dutify-cloud-ai-skill.git <your-project>/.claude/skills/dutify-api

Verify

After install:

ls ~/.claude/skills/dutify-api/SKILL.md          # should exist
ls ~/.claude/skills/dutify-api/references/       # product references

Then in any Claude Code session, ask "what's the URL for the Dutify task-type catalogue endpoint?" — Claude should pick up the skill, load references/task-types.md, and answer.

How the skill is laid out

The skill is topic-indexed rather than a single big document. SKILL.md is the orientation file an LLM always sees; the product reference files in references/ are loaded on-demand based on the topic map.

File	Topic
SKILL.md	All-product routing and topic map; Suite conventions remain in references/suite-guide.md
references/auth.md	API key header, the 40+ scopes, bound workspace via /v1/api-keys/current
references/errors.md	PM nested vs FR/Wiki flat envelopes, error-code vocabulary, rate-limit response, network errors
references/tasks.md	/lite/context, search, create, update, comments, relationships, attachments, recurrence, time entries, Idempotency-Key, non-lite TaskCreationRequest, cross-workspace move
references/task-types.md	Catalogue, the 9 SYSTEM rows, cascade default (3 levels — list→folder→space, no workspace fallback), single + bulk type changes, default-at-level setters, reassignment-on-delete
references/wiki.md	Wiki pages list/read/write, Markdown vs TipTap body formats, search query syntax
references/roadmarq.md	Feature requests + bugs, per-board short-ID prefixes, lite verbs (/votes not /vote, /moderation not /moderate), ?workspaceIdentifier= query rule
references/webhooks.md	Subscriptions, payload envelope, X-Webhook-Signature-256, retry schedule, WebSocket ticket flow, activity log read shape
references/custom-fields.md	Custom-field CRUD (lite), value-setting on tasks, formula configuration (structured JSON), rollup configuration
references/views.md	Views (Lite): create/list/get/update/clone, column visibility/order, filters, MCP create behavior
references/dashboards-forms.md	/v1/dashboard/lite, form admin (config + submissions + CSV export), public form submission
references/sprints.md	Sprint groups, sprint lifecycle (start/complete/rollover), membership, burndown — non-lite
references/imports.md	CSV import: preview → confirm flow, one-shot upload, polling job status
references/admin.md	Plan-quota errors (402 QUOTA_EXCEEDED), status / priority / level CRUD per scope, teams, whiteboard collab token
references/members.md	Workspace member list / role / remove, invitations send / cancel / accept / decline
references/notifications.md	List + count + mark-read, page-style pagination quirk, data field is a real JSON object (unlike activity log)
references/filters.md	Saved filters per list, per user, numeric Long id, no "apply saved filter" endpoint on lite search

Authentication

Data-access calls use X-API-Key: du_live_… for granted products, dk_live_… for a Suite workspace, or dh_live_… for a Hub workspace. Create workspace keys in workspace settings and personal keys in account settings. Workspace keys retain their fixed binding; personal keys discover allowed workspaces and select one per call with X-Dutify-Workspace. See authentication and personal keys for scopes and denial codes.

Why a topic-indexed skill rather than one long doc

Skills load fully whenever they're triggered — a 1500-line single doc would burn that much context per task. Splitting orientation in SKILL.md and detail per topic means a wiki-only task only loads wiki.md; a webhook setup loads webhooks.md + auth.md; nothing else. The entry point is concise; each reference averages ~95 lines.

Evaluation

evals/evals.json carries 5 prompts that exercise the skill end-to-end:

• create-task-python — production-quality Python script that creates a task by name
• list-wiki-pages-curl — curl + a notes file
• update-fr-status — a single-shot HTTP request to change a feature-request status
• bulk-retype-tasks — Python that finds all tasks in a list, looks up the workspace's task-type identifiers, and bulk-changes their type via the /v1/tasks/bulk/change-type endpoint
• find-untyped-tasks — curl + notes explaining the cascade-default chain and how taskType: null arises

These are intended for evaluation harnesses (e.g. Anthropic's skills-eval framework) that score with_skill vs without_skill runs against expectation lists.

Contributing

The canonical source for this skill lives at Dutify-suite/skills/dutify-api/ inside the Dutify suite monorepo; this GitHub mirror is for distribution. Edit at the source and re-publish; don't open PRs against this repo unless you're publishing a fix that's already merged upstream.

License

Internal Dutify documentation. Use of the API requires a valid Dutify API key.

Maintaining the Hub compatibility distribution

Canonical Hub references live in references/hub. Check copies with python scripts/sync_hub_references.py --hub-skill <hub-repo>; use --write to export manifest files. The Hub distribution ships complete references and never requires the all-product skill at runtime. Update both distributions and their version files together when shared behavior changes.