Once your AI client is connected, you talk to your work in plain language. This page shows what to ask, grouped by the job in front of you, then a worked example of the assistant planning a sprint end to end, and prompts by role. For the tools behind the prompts, see the Tool Reference.
Begin with a few fast, safe, read-only prompts to build confidence:
You can also just ask "what can you do?" — the assistant has a built-in catalogue.
Even when there is no purpose-built tool for what you ask, the assistant can reach the DevStride API directly through invoke_api_operation, so less common administrative and configuration tasks are within reach too — always within the permissions you already have.
The assistant can set up dependencies and move dates, so you can keep a schedule current in plain language:
If your organization has Link Mode turned on, moving an item's due date through the MCP automatically adjusts the items that depend on it — the same behavior you get on the Gantt chart. See Link Mode for how dependent dates update.
The pattern to copy: the assistant analyzes and proposes, you approve, and only then does it change anything. Point it at a sprint's worth of unestimated work:
The result is a fully structured, dated roadmap built from your real backlog — drafted by the assistant, reviewed and approved by you at every step.
The same connection serves very different jobs. A few starting points:
Project / delivery manager
Team member
Service delivery leader
Executive (CTO / VP)
create_workstream, update_workstream, archive_workstream, unarchive_workstream and delete_workstream, with workstream* keys in their inputs and results. The older *_folder tools, and the *_lane tools for statuses, were removed on October 10, 2026. Switch saved prompts, scripts, and custom tooling to the new names — see API keys.update_item (or to PATCH /v1/organizations/{organizationId}/work-items/{itemNumber}) is refused straight away with a 400 that names the workstream route; use update_workstream (PATCH …/workstreams/{itemNumber}) instead. Before September 2026 that mistake hung for about 30 seconds and came back as a server error.Your assistant can search across sub-tasks, list the sub-tasks of a parent, and create, update, or delete them. Item searches can also include items whose sub-task is assigned to you — so you can ask for "everything I'm working on, even via sub-tasks." Results are always filtered to what you have permission to read.
Personas are available through the API and MCP alongside person assignments. For example:
The same permissions and person/persona compatibility rules apply as in the app. A persona never grants item access. Catalog updates use the last-read catalog revision to prevent one administrator overwriting another's changes; saved personas are retired rather than deleted. See Managing personas.
An item type can be allowed under more than one parent type, for example a Story under both a Feature and an Epic. The assistant reads the full set before it creates or moves anything:
get_work_type_hierarchy lists, for each type in byName, every permitted parent in parentTitles and parentWorkTypeIds (the primary parent first), its childrenTitles, and ancestorChains, one top-to-type chain for each route through the hierarchy. Its tree is a simpler picture of the same hierarchy, so read byName whenever a type has more than one parent.find_parent_candidates returns candidate parents of every permitted parent type in one call, and lists those types in parentWorkTypes.parentTitle is now parentTitles, ancestorChain is now ancestorChains, and parentWorkType is now parentWorkTypes. If you have scripts or saved prompts that read the old names, update them. Through the REST API, a work type's permitted parents are permittedParentWorkTypeIds.An item's type has to fit its parent and its children, so the assistant checks before it changes anything:
preview_work_type_compatibility answers, for one item, which types it could change to on its own. For each candidate it says whether the item would still fit under its parent and whether every direct child would still fit under it, naming the pairs that would break.preview_convert_item_tree is a dry run of converting a parent and everything beneath it with a mapping such as "Epic becomes Feature, Story becomes Epic". It reports how many items the tree holds, every parent/child pair the mapping would break, and any item whose board has no status for its new type. It changes nothing.convert_item_tree applies that mapping to the whole tree. Either every item converts or none does, and a tree larger than 200 items (the parent and archived items included) is refused without changing anything. You need edit access to every item in the tree.A good prompt: "Preview converting I1234 and its children so the Epics become Features and the Stories become Epics. Show me what would break, and wait for my approval." The same conversion is available in the app — see Changing an item's type.
Your assistant can work the Knowledge Base directly: find and search articles, read and write their content, move them through their lifecycle (publish, unpublish, archive, unarchive), set which companies see them, and manage the category tree.
It can also read an article's version history — who changed it and when — and restore an earlier version. A restore changes only the title and body, adds a new entry to the history, and deletes nothing, exactly as it does in the app. See Article history and restore.
The assistant can take part in an article's comment threads, the same threads you see in the editor's Comments panel:
| Tool | What it does |
|---|---|
list_kb_article_comments | Lists an article's threads, oldest or newest first, open or resolved only, with the words each pinned thread is about |
add_kb_article_comment | Starts a new thread, such as a review note, on the whole article or pinned to a passage of it |
reply_to_kb_article_comment | Replies inside a thread |
resolve_kb_article_comment / unresolve_kb_article_comment | Marks a thread resolved, or reopens it |
set_kb_article_comments_enabled | Switches the article's customer comments on or off |
Comments the assistant writes are internal notes unless you explicitly ask for a public one. A public comment appears only in the customer thread on the portal, under the article, and only while customer comments are on for it. A public reply in a customer's thread is also emailed to the customers taking part, who can answer by email; a new thread is not. See Email replies to customers.
The assistant can also pin a new thread to the words it is about, just as you do by selecting words in the editor before you comment: ask "comment on the sentence about refunds in this article" and it quotes those words when it posts. When it reads a thread, it sees which words the thread is pinned to and whether they are still in the article or the passage has been removed, so it can tell you what each comment was about even after the text changed. If the words it quotes are no longer in the article, nothing is posted, and it reads the article again before retrying. When it edits an article's text, it keeps the pins, and DevStride puts back any it leaves out wherever their words are unchanged.
Article summaries also say whether customer comments are on and how many review threads are still open, so "which articles have unresolved feedback?" is a single question. See Article review comments and Customer comments on articles.
Setting Up Your Connection
Connect your AI client to the DevStride MCP Server with OAuth 2.1 or an API key, using the ready-to-paste snippet for Claude, ChatGPT, Cursor, Copilot, Codex, or Gemini CLI.
Best Practices & Tips
Habits that make the DevStride MCP Server fast, accurate, and safe: start read-only, ask it to propose before it changes anything, refer to things by name, and keep the connection healthy.