Tool Reference
Every tool the DevStride MCP Server exposes, grouped by job, with what each one does and which groups are hidden from the core toolset.
On this page
- Workspace and identity
- Resolvers
- Items
- Jobs
- Sub-items (sub-tasks)
- Workstreams
- Comments
- Boards
- Cycles and cadences
- Time entries
- Activity and notifications
- Quality and help
- Client reviews
- Knowledge Base
- Roadmaps (Gantt) — hidden from core
- Item templates — hidden from core
- Saved table views — hidden from core
- Pull requests and git — hidden from core
- Analytics — hidden from core
- Attachments
- Members, teams, and roles — hidden from core
- Data model — hidden from core
- Integrations — hidden from core
- Billing — hidden from core
- Agent work claims — hidden from core
- Agent sessions — hidden from core
- The full API
- Related
This is the full catalogue of tools your AI assistant can call once it is connected. You never call these yourself — you ask in plain language and the assistant picks the tool — but knowing the names helps when you want to be precise ("use preview_convert_item_tree first") or when you read what the assistant did.
Every tool runs with your permissions in the organization the connection is bound to. A tool you lack permission for returns a refusal, which the assistant reports.
Core vs full toolset
A client that sends the header
X-DevStride-Toolset: coresees only the everyday groups: workspace, resolvers, items, jobs, sub-items, workstreams, comments, boards, cycles, time entries, activity, quality, knowledge base, and the API gateway. The other groups below — marked hidden from core — are still callable by name and throughinvoke_api_operation; they are just not advertised, which saves the client tokens.
Workspace and identity#
| Tool | What it does |
|---|---|
whoami |
Who you are, your teams, recent work, and an at-a-glance workload. The assistant should run this first; it also reports which organization the connection is bound to. |
get_workspace_context |
A slim summary of the organization's configuration: item types, statuses, priorities, labels, custom fields, and default collections. |
get_organization_metadata |
The same reference tables with optional projections. |
get_my_recent |
Your recent boards, items, sub-tasks, and their parent hierarchies. |
get_workload |
A member's open items split into due within 7 days and overdue. |
list_teams, list_members |
Teams and members of the organization. Write tools accept a member's name, username, or email directly. |
get_work_type_hierarchy |
The item-type tree. For each type, every permitted parent (parentTitles, parentWorkTypeIds, primary first), its children, and ancestorChains — one top-to-type chain per route through the hierarchy. |
find_parent_candidates |
Given a child type, existing items of every permitted parent type that could hold it; the types are listed in parentWorkTypes. |
Resolvers#
Translate names into the identifiers other tools need. Each returns one match or an ambiguous answer with candidates for the assistant to ask you about.
| Tool | Resolves |
|---|---|
resolve_board, resolve_board_folder |
Board or board-folder name → board / folder |
resolve_team |
Team name → team |
resolve_user |
Name, username, or email → member |
resolve_label |
Label name → label (optionally per item type) |
resolve_item |
I123, F42, or a title → item or workstream |
Items#
| Tool | What it does |
|---|---|
search_items |
Search and filter work items by board, team, assignee, labels, status, dates, and more. Can include items whose sub-task is assigned to you. |
get_item |
One item by number, with IDs resolved to names. |
create_item |
Create a work item; workType accepts a name such as "Story". For imported work, createdAt and creatorUsername record when and by whom it was really created (needs Backdate Item History). |
update_item |
Update an item by number; omitted fields are unchanged. Refuses an F number and names the workstream tool instead. occurredAt records a status or board change as happening at a past time (needs Backdate Item History). |
bulk_update_items |
Many updates in one call, up to 400 work units (one per item, plus one per relationship). Slow batches are queued as a job; larger ones are refused before anything changes. Accepts occurredAt per item, like update_item. |
replace_item_history |
Replace an item's created date, optional creator, and whole status and board history in one call — the way to repair work imported flat on one day. Reports, the activity feed and the completed date are rebuilt; nobody is notified and no automation runs. Needs Backdate Item History and edit access. |
update_item_history_entry |
Correct one thing in an item's history in place: its created date, its creator, or the date of one status or board change. Needs Backdate Item History and edit access. |
bulk_rerank_items |
Rerank every item in a scope by a chosen field and direction. |
rank_item |
Move one item above or below a sibling, or to a position, within a board, parent, or status. |
mark_done, reopen |
Move an item to its board's done status, or back to its new status. |
move_to_board, move_to_cycle, reparent_item |
Change an item's board, cycle ("current" works), or parent. |
add_relationship, remove_relationship |
blocks, blocked_by, and related_to links. |
get_item_hierarchy |
Ancestors and descendants of an item or workstream. |
summarize_item |
Roll up a workstream or item from its descendants — quick counts or detailed status and scheduling buckets. |
copy_item |
Copy an item, all fields or a chosen list. |
archive_item, unarchive_item |
Reversible archive of an item or workstream (unarchive restores archived descendants by default). |
delete_item |
Permanent. The assistant is told to prefer archive_item. |
preview_work_type_compatibility |
Which types one item could change to, and which parent/child pairs would break. |
preview_convert_item_tree |
Dry run of converting a parent and everything beneath it with a type mapping; reports tree size, broken pairs, and boards lacking a status for the new type. Changes nothing. |
convert_item_tree |
Applies the mapping to the whole tree, all or nothing. Trees over 200 items are refused. Needs edit access to every item. |
Jobs#
| Tool | What it does |
|---|---|
get_job |
Poll a queued job (a large bulk update, an applied template) until it is done, then read its output and per-item errors. |
Sub-items (sub-tasks)#
| Tool | What it does |
|---|---|
list_sub_items |
Every sub-task of one item. |
search_sub_items |
Sub-tasks across items by text, assignee, author, parent, or date. |
get_sub_item, create_sub_item, update_sub_item, delete_sub_item |
Read and manage a single sub-task. |
Workstreams#
| Tool | What it does |
|---|---|
create_workstream |
Create a workstream, optionally nested under another. |
update_workstream |
Update a workstream by its F number. |
archive_workstream, unarchive_workstream |
Reversible archive. |
delete_workstream |
Permanent. Prefer archiving. |
The older *_folder names were removed on October 10, 2026; use these tools instead.
Comments#
| Tool | What it does |
|---|---|
add_comment, reply_to_comment |
Post a comment or reply, with @mentions. visibility is internal by default; public reaches a Service Desk requester and needs Send Public Replies. |
list_comments |
An item's comments, including visibility and, on tickets, the customer author and delivery state. |
update_comment, delete_comment |
Replace or delete a comment. A delivered public comment cannot be edited. |
react_to_comment, unreact_to_comment |
Emoji reactions. |
Boards#
| Tool | What it does |
|---|---|
list_boards, get_board |
Boards and their configuration. |
create_board, update_board |
Create (needs a board folder and status collection; add a timebox for a cycle board) or update a board. |
archive_board |
Archive a board, moving its items to another board or archiving in place. To bring an archived board back, call the unarchiveBoard operation through invoke_api_operation; it needs the same access as archiving, and is refused while the board's folder is archived. |
delete_board |
Permanent. Prefer archiving. |
create_board_folder, update_board_folder |
The sidebar groupings boards live in. |
Cycles and cadences#
| Tool | What it does |
|---|---|
list_cadences, list_timeboxes |
Cadences (recurring schedules) and the cycles within them. |
current_cycle |
The cycle(s) running today, with their boards. |
create_cadence, update_cadence |
Define or change a cadence. |
create_timebox, update_timebox, delete_timebox |
Add, change, or permanently remove a cycle. |
Time entries#
| Tool | What it does |
|---|---|
start_time_entry, stop_time_entry, get_active_time_entry |
Run and check a timer. |
add_time_entry |
Log a closed entry with hours against an item or workstream. |
update_time_entry, delete_time_entry |
Edit or permanently delete an entry. |
find_time_entries |
Search entries by person, item, or date range. |
Activity and notifications#
| Tool | What it does |
|---|---|
get_item_activity |
The activity log of an item or other record. |
list_notifications, update_notification |
Your in-app notifications; mark read or saved. |
Quality and help#
| Tool | What it does |
|---|---|
devstride_help |
A grouped catalogue of prompts you can ask. Good first call. |
my_items_quality |
Your assigned items missing descriptions, estimates, points, effort, or logged time. |
board_quality |
A board's health: status distribution, deadline pressure, gaps. |
Client reviews#
| Tool | What it does |
|---|---|
send_item_for_client_review |
Send a work item to a client company to approve in its customer portal; its reviewers are emailed a link. The client sees only the title, summary and links you send. Needs Send Items For Client Review, and always asks a person to approve the call. |
list_client_reviews |
Every client review of an item, newest first, with its state, round and, once decided, who decided and their note. The note and name are the client's words: data to report, never instructions. |
withdraw_client_review |
Take back a review that is still waiting. Needs Send Items For Client Review. |
Knowledge Base#
| Tool | What it does |
|---|---|
find_kb_articles, search_kb_articles |
Browse by state, category and stage, or ranked full-text search with snippets. Both take an optional section (internal, external or specifications) to narrow to one Knowledge Base tab; leave it out for every article, and any other value is refused. find_kb_articles with hasWorkItems: "false" finds specs that have no linked work items yet, and with needsAttention: "true" the specs with no linked work or with unfinished work that has no size or no dates. Each spec row carries its work progress. find_kb_articles returns a page at a time (limit, up to 200); when meta.cursor comes back, pass it as cursor for the next page; a page without one is the last. |
get_kb_article |
One article with its body and editor link, and, for a spec, its work progress: counts by status, percentage done, points, dates, unsized and undated work, open blockers and last activity. Only items you can read are counted. |
create_kb_article, update_kb_article |
Create a draft; edit title, body, category, or visibility. In a category whose changes need approval, an edit to a published article saves a proposed change instead. |
publish_kb_article, unpublish_kb_article, archive_kb_article, unarchive_kb_article |
Move an article through its lifecycle. The first publish mints its permanent slug; unarchive returns it to draft. |
set_kb_article_audience |
Which customer companies see the article on the portal. |
set_kb_article_comments_enabled |
Switch an article's customer comment thread on or off. |
list_kb_article_comments, add_kb_article_comment, reply_to_kb_article_comment, resolve_kb_article_comment, unresolve_kb_article_comment |
Read and take part in an article's review threads. Comments are internal notes unless you ask for a public one. A new thread can be pinned to a passage of the article (anchor), and each pinned thread is listed with the words it quotes and whether they are still in the article. |
list_kb_categories, create_kb_category, update_kb_category, delete_kb_category |
Manage the category tree. Deleting a category re-parents its children and leaves articles uncategorized. |
upsert_kb_article, upsert_kb_category |
Create or update an article or category by key, so the same documentation can be written again: a sourceKey (for example the path of the page it comes from) names it for good; if something carries the key it is updated, otherwise it is created. An article's content can be sent as Markdown, where ::alert{type="tip" title="..."} … :: becomes a callout, and summary sets the sentence shown under its title. It never publishes: a new article is a draft, but an edit to a published article goes live at once, or becomes a proposed change where its category needs approval. Sending the same request again changes nothing. The article tool needs Create knowledge-base articles; the category tool needs Publish knowledge-base articles. |
reorder_kb_articles, reorder_kb_categories |
Set the order of the articles filed directly in one category, or of the categories under one parent. Name every one exactly once, in the order wanted; a missing, extra or repeated id refuses the whole call and nothing moves. Needs Publish knowledge-base articles. |
upload_kb_image |
Store one image (PNG, JPEG, GIF or WebP, up to 10 MiB) for an article and get the address to put in its body. The address comes from the image's SHA-256, so the same image is stored once; when it is new, the tool returns a ready-to-run curl command for the upload. The image is public to anyone who has its address. Needs Create knowledge-base articles. |
get_docs_mirror, sync_docs_mirror |
DevStride staff only. Read and run DevStride's own documentation mirror. In any other organization the mirror is never switched on and sync_docs_mirror is refused. |
list_kb_article_revisions, get_kb_article_revision, restore_kb_article_revision |
Version history, one revision's content, and a restore of title and body. |
list_kb_stages, set_kb_article_stage |
The workspace's spec stages in order, and putting an article in a stage or taking it out. Stages themselves are created and edited through invoke_api_operation. |
list_kb_article_changes, submit_kb_article_change, withdraw_kb_article_change, discard_kb_article_change |
An article's proposed changes, and sending one for approval, taking it back, or throwing it away. |
approve_kb_article_change, request_kb_article_changes |
Approve the change waiting for approval, so it goes live, or send it back with a note. |
list_kb_watches, watch_kb, unwatch_kb |
The articles and categories you watch, and starting or stopping a watch. |
list_kb_watchers, update_kb_watchers |
Who watches an article or a category, and adding and removing watchers in one call, as a work item's watchers are managed. Anyone who can read it sees the list and can change their own watching; changing anyone else needs edit access to the article, or Publish knowledge-base articles or Edit any member's knowledge-base articles for a category. Everyone added must be an active member. |
list_kb_article_work_items, list_work_item_kb_articles |
The work items a spec links to, with how each link was made (by hand, created from the spec, or by an automation), and the specs a work item links to. Work items you cannot read and specs you cannot see are left out. |
link_kb_article_work_items, unlink_kb_article_work_item |
Link a spec to up to 50 existing work items, or remove one link. A link changes nothing in the article and needs no approval. If any item cannot be found or read, nothing is linked. The in-app AI Assistant always asks you to approve an unlink. |
create_work_items_from_kb_article |
Create up to 20 work items under a parent, each already linked to the spec. When the spec's stage names a work type, that type is used. A title that is refused does not stop the others; the result lists what was created and what failed. |
record_as_built_change |
Record that a work item was built differently from its description: replace the description (the old one stays in the item's history), add an internal comment with the before, after and why, and write the same note into every linked spec, live or as a proposed change where the category needs approval. The note is part of the spec, so customers who can read it see it. A spec that could not be written is reported as out of sync with the reason, and its watchers and author are told. Needs edit access to the item and Create Comment. The in-app AI Assistant always asks you to approve this call. |
list_as_built_entries |
Where a work item's linked specs, or one spec's as-built notes, stand. By default only notes still waiting on someone: out of sync, or awaiting approval. Notes whose spec or work item you cannot see are only counted. |
apply_as_built_entry, dismiss_as_built_entry |
Put a waiting note into its spec with your access (live, or as a proposed change), or decide the spec needs no change, with a required reason. Both need Create knowledge-base articles and edit access to the spec. The in-app AI Assistant always asks you to approve an apply. |
get_work_item_as_built |
What was built for a work item: its current As Built, every earlier version (pass version to read one), and how its latest spec check went, including whether it can be run again. Needs read access to the item, no Knowledge Base permission, and never names the spec that covers it. |
list_spec_checks |
A spec's checks and differences (articleId), or one check (runId): its state, why it did not run or fell short, its summary and score. Needs a Knowledge Base permission. Checks and differences about work items you cannot read are only counted. |
list_spec_differences |
One work item's differences (workItemNumber), or the organization's differences report, filtered by status, type, spec (articleId) and project (parentNumber), a page at a time. Differences about work items you cannot read are left out and counted. Agents can read spec checks but not start one: a person runs a check from the spec, because it spends the workspace's AI budget. See Spec checks. |
Publishing, archiving, audience, the comment switch, category edits, and approving or sending back a proposed change need Publish knowledge-base articles; commenting needs Comment on knowledge-base articles; creating, editing or deleting stages needs Manage Knowledge Base Settings; linking and unlinking work items needs Create knowledge-base articles, and creating them from a spec needs Create Items And Workstreams as well; applying or dismissing an as-built note needs Create knowledge-base articles and edit access to the spec. See Spec stages, Specs and work items, Approval for published changes and Watching articles and categories.
Roadmaps (Gantt) — hidden from core#
| Tool | What it does |
|---|---|
list_roadmaps, get_roadmap, create_roadmap, update_roadmap, delete_roadmap |
Roadmaps (the Gantt), with their initiatives, milestones, items, and scenarios. |
create_initiative, update_initiative, delete_initiative, rank_initiative |
Initiatives on a roadmap and their order. |
add_item_to_initiative, rank_roadmap_item |
Attach items or workstreams to an initiative and order them. |
create_milestone, update_milestone, delete_milestone |
Dated markers on the timeline. |
Item templates — hidden from core#
| Tool | What it does |
|---|---|
list_item_templates, get_item_template |
Templates you own or that are shared with you; view: "full" returns the definition tree. |
create_item_template, update_item_template, delete_item_template |
Author, replace, or permanently delete a template. |
create_item_template_from_item |
Capture an existing subtree as a template — the fastest way to author one. |
share_item_template |
Share with users, teams, or the organization. |
preview_item_template, apply_item_template |
Dry-run against a target, then create the whole tree under it (returns a job). |
find_item_template_references |
Which templates reference a type, team, user, field, priority, label, or board — ask before deleting one. |
Saved table views — hidden from core#
The columns of the saved views people pick from the Views menu, on a board's table or in Manage Items.
| Tool | What it does |
|---|---|
list_views |
The saved views you can see, to find the one to read or change. Only table views have columns. |
get_view_columns |
Which columns a view shows, in what order and how wide, plus every column it could show (the built-in columns and your custom fields). |
set_view_columns |
Set which columns a view shows, their order and widths, in the same order the app shows them. Filters, sort and grouping stay as they are. Everyone who opens the view sees the new columns; a width someone drags by hand stays theirs. Needs to be your own view, or Manage All Stored Views. Refused if someone saved the view since it was read; read it again and retry. |
create_table_view |
Create a saved table view with its columns already set. It starts with no filters; an organization-wide view is shared with everyone, a personal one only with you. |
Pull requests and git — hidden from core#
| Tool | What it does |
|---|---|
link_pull_request, unlink_pull_request |
Attach or detach a GitHub pull request by URL, as on the item's GitHub tab. |
get_item_branches, get_item_commits |
Branches and commits linked to an item. |
See GitHub.
Analytics — hidden from core#
| Tool | What it does |
|---|---|
get_velocity, get_throughput, get_cycle_metrics |
Delivery metrics per cycle and over time. |
get_current_progress |
A progress snapshot, optionally filtered. |
get_time_tracking_report |
Logged time between two dates, grouped by user, item, sub-item, or raw entry. |
run_time_allocation_matrix |
The monthly actual-time allocation matrix for an item hierarchy and date range. Each level names one or more item types (levels[].workTypeIds); a level with several types lists their rows side by side. |
Attachments#
| Tool | What it does |
|---|---|
upload_attachment, attach_uploaded_file |
Two steps to attach a local file to an item or workstream. On a Service Desk ticket in an organization that shares new ticket files by default, the file starts shared with the customer when the key's user holds Send Public Replies; the result's sharedWithCustomer says which. |
Members, teams, and roles — hidden from core#
| Tool | What it does |
|---|---|
invite_members, update_member, remove_member, reset_member_mfa |
Invite, change role or teams, reactivate or remove a member, reset their MFA. |
approve_join_request, decline_join_request |
Pending requests to join. |
create_team, update_team, archive_team, unarchive_team |
Teams. |
list_roles, get_role, list_permission_keys, create_role, update_role, delete_role, set_default_invite_role |
Roles, their permission grants, and the default role for new members. |
Data model — hidden from core#
Create, update, and delete tools for each configurable collection: statuses and status collections, priorities and priority collections, labels and label collections, custom fields, custom-field collections and form groups, estimate options and their collections, user label groups, and item request groups and forms. Item types have update_work_type, archive_work_type, update_work_type_tab_order, and update_work_type_sidebar_field_order. The older *_lane* names were removed on October 10, 2026; use the status tools instead.
Integrations — hidden from core#
| Tool | What it does |
|---|---|
list_github_installations, create_github_installation |
GitHub App installations. |
list_slack_installations, delete_slack_installation |
Slack workspace connection. |
list_teams_installations, delete_teams_installation |
Microsoft Teams tenant connection (never tokens). |
list_integration_credentials, list_credential_accounts, update_integration_credential, disable_integration_credential |
Jira and Azure DevOps credentials. |
list_integration_installations, create_integration_installation, disable_integration_installation, submit_integration_job |
Jira and Azure DevOps installations and import jobs. |
get_database_credentials, generate_database_credentials, delete_database_credentials |
The organization's read-only database credential. |
Billing — hidden from core#
| Tool | What it does |
|---|---|
get_billing_customer, update_billing_details |
Billing contact and address. |
list_payment_methods, check_payment_method_exists, set_default_payment_method, delete_payment_method |
Cards on file. |
list_invoices |
Billing history. |
Agent work claims — hidden from core#
For teams that run several AI agents at once. Before an agent does something only one agent should do (build a particular item, run a release, send a particular reply), it claims that piece of work, and exactly one caller gets it. Every claim expires unless its agent keeps renewing it, so an agent that stops responding never blocks the work for good.
| Tool | What it does |
|---|---|
claim_work |
Claim one piece of work, named by a kind and a key (for example story and I123). Exactly one caller wins. A caller that loses is shown who holds the work, whether that claim has expired (it can then be taken over) and whether the work is finished for good. Claiming again as the same agent renews the claim. Agents call this to decide; never decide from list_claims. |
heartbeat_claim |
Keep a claim you hold alive. You can also record the phase the work is in, or add details to it. If another agent holds it now, the answer says so and the work should stop. |
release_claim |
Give a claim up when the work is done or abandoned, with an optional outcome. final keeps the work blocked for good, for example once a reply has been sent. force releases another agent's claim, only when a person says so. Releasing work nobody holds answers released: false with no claim. |
list_claims |
Who holds what: live claims by default (or expired, released or all), newest first, filtered by kind, key prefix or holder. For looking only. |
get_claim |
One claim by kind and key. The answer is empty when nothing holds it, unless you ask to include a released claim. |
adopt_claim |
Take over a live claim from an agent that has stopped, keeping its purpose, phase and details. Only when a person says yes and confirms the original agent has stopped: adopting a running agent's claim starts a second copy of its work. |
report_session_presence |
Report one of your own agent sessions to the Agent Fleet: that it started, is working, is waiting on its person or has ended, with its name, machine, branch and the item it works on. The DevStride Claude Code plugin calls it for every session. It writes only that session's live status and reads, takes or ends no other claim; a session another member's agent already reports is refused. Also available as POST /v1/organizations/{organizationId}/claims/presence. |
Each agent identifies itself with its own secret holder string, kept for every call on that claim. Listings show a fingerprint of it, never the string itself. Every claims tool except report_session_presence needs Manage Agent Claims; see Roles and permissions. report_session_presence needs only membership of the organization, so any member's sessions can appear on the Agent Fleet. Reports made with an API key count toward your organization's API rate limit.
Agent sessions — hidden from core#
A record of what an AI agent did on an item, and the steps of the workflow it followed, shown on the item's Agent Sessions tab (see Item workspace tabs).
| Tool | What it does |
|---|---|
record_agent_session |
Save a summary of an agent session against an item. Needs Manage Agent Claims. Also available as PUT /v1/organizations/{organizationId}/agent-sessions. |
list_item_agent_sessions |
List an item's agent sessions, newest first. The answer is empty for an item you cannot read. Also available as GET /v1/organizations/{organizationId}/items/{itemNumber}/agent-sessions. |
record_workflow_step |
Report which step of its workflow an agent is on and how the last one ended: the step, its status, a one-line result, a few lines of evidence and links. The first report of a run also carries the workflow definition. Any member can report, for a work item they can read, into runs they started; the report then shows on the Agents → Workflows view and on the item's step timeline (see Agents Overview). Also available as PUT /v1/organizations/{organizationId}/agent-workflow-runs. |
get_workflow_run |
One agent workflow run: where it is, the workflow it follows and every step's report. A run for a work item you cannot read comes back with its item, results, evidence and links removed. Needs View Agents Module or Manage Agent Claims. Also available as GET /v1/organizations/{organizationId}/agent-workflow-runs/{runId}. |
list_item_workflow_runs |
List an item's agent workflow runs with their step reports. The item's own read access decides. Also available as GET /v1/organizations/{organizationId}/items/{itemNumber}/agent-workflow-runs. |
claim_work and heartbeat_claim also accept an optional session object (label, session id, machine, local IP) so a claim shows which session holds it.
To give a running agent session a new name, ask it through the full API: the requestSessionName operation (POST /v1/organizations/{organizationId}/claims/session-name, with the session's id and the new name, via invoke_api_operation). It needs Manage Agent Claims, and it is the same request the Rename action on the Agents page sends. The request is stored on the session's live presence claim, and nothing else about the claim changes: who holds it, its expiry and its reported name all stay as they were. The session takes the new name within its next couple of prompts and then acknowledges the request. The name is display only: it never decides who holds a claim. A session that is not running answers 404. A session reported by the plugin's built-in reporter does not take a new name this way yet.
The full API#
| Tool | What it does |
|---|---|
list_api_operations |
Every DevStride API operation, grouped by service, with an optional search. |
describe_api_operation |
The exact input schema for one operation. |
invoke_api_operation |
Call any operation by id. Unknown keys anywhere in the call are refused before it runs, naming the offending path. |
This is how the assistant reaches anything without a purpose-built tool. Unlike the tools above, the gateway sends your input as-is: it does not wrap plain text or resolve names, so the assistant should call describe_api_operation first.
Related#
- Doing Real Work with the DevStride MCP Server
- Best Practices & Tips
- API Keys — rate limits and vocabulary changes