On this page
Dependencies
Dependencies shows one element's dependencies in three ways: what depends on it, what it depends on, and the route from it to another element. An element is a node, table, endpoint, API client, channel or page.
Each type follows dependencies hop by hop through your boards and their aspects as they stand, across layers and apps. It follows the edges drawn on the diagram, and also what the code proves but no board draws, such as a client calling an endpoint. The same boards always give the same result.
A route that matters can be saved as an insight.
Three ways to use it
Open Dependencies
Open Dependencies from an element. On a board you can edit:
- From the node. Right-click a node, point to Dependencies, and choose Impact, Depends on or Route.
- From the inspector. Select a node. The inspector's Dependencies section lists the three types; choose one.
- 1
The options, folded
The options fold to one line once the result is drawn; Edit opens them again.
- 2
The selected element's route
Select an element and the panel shows the route that reaches it as a row of stops.
- 3
What the canvas shows
A label at the top of the canvas names what is shown and counts the named nodes on the board in view.
The panel opens on that node, over the whole tree the board belongs to. Route starts from the node and waits for you to name To. The panel takes the same column as the Insights Bar, so opening one closes the other.
Change what it shows
- Choose a Type.
- Check Scope. It starts at the top board, which covers every board beneath it; pick a lower board to stay inside one app.
- Name the Element, or From and To for a Route; Swap reverses them. Type a name, or select the target button beside the field and click a node. A table, endpoint, API client, channel or page has no node, so search for it.
The result is drawn as soon as the element is named, and the options fold to one line; Edit opens them again.
- Depth is how many hops to follow, from 1 to 6; it starts at 3. Each depth shows how many elements it reaches, and Go one hop deeper appears when more lies beyond.
- Follow chooses which kinds of dependency count: Drawn connections, Data, APIs, Events, Layers and Structure. Structure starts off for every type, and Layers starts off for Depends on.
Route has no Depth: it finds the shortest route, up to 6 hops, from From to what From depends on. If there is none, try Swap.
Read the result
The canvas dims every node the result doesn't name; close the label at the top of the canvas to stop. In the panel, each result is a Picture or a Table. A circle is a node, a square is an aspect row, and a hollow mark or dashed line is not drawn on any board: only code proves it.
Follow an element
Point at an element in the picture to light its route back to the start. Select it, and the canvas moves to it, opening its board if needed, and dims everything off that route. The panel shows the route as a row of stops, with the evidence when code proves the last hop. Select a stop to move the canvas there, or Open the table (or endpoint, client, channel, page) to see that aspect row on its own screen.
Narrow the result
Under the picture or table, Where, What and Reached by break the elements down by system, kind, and whether the last hop is drawn. Select a bar to filter; filters combine, and each shows as a chip you can remove.
Save a route as an insight
Save as insight keeps one route as an insight on the board where the route starts. It appears on a selected element, and under a Route result. View, on the confirmation, opens the new insight.
The insight's trail is the route, so it reads on the canvas step by step like any other insight. It is saved as an observation you wrote: approved on arrival, in no batch, and editable from the Insights Bar.
A saved insight is the route as it stood when you saved it. To see the route on the boards as they stand now, open the insight and select Ask this again in its header.
A whole impact or dependency result can't be saved as one insight. Save the routes that matter.
Ask in Chat
Open Chat and ask in your own words, such as what breaks if the catalog database changes. The reply carries a dependency card for the type and element you meant, up to three per reply. Dependency cards covers what a card shows.
With a node selected, Chat's quick starts include What depends on {name}? and What does {name} depend on?; with two selected, How does {first} reach {second}?. A Chat turn uses credits like any other.
Use it from an agent
Over MCP access, an AI host, a coding agent, a review bot or a pipeline can use the same three types. Each type is one tool, one call over a board's whole tree, and all three need only the Read scope.
With the workspace plugin
The workspace plugin's /ask-board turns a request in your own words into the right call, and lays the result out in your AI host.
The result opens with a headline and the spread across systems and hops, then a short summary of the facts. A picture follows, drawn for the type: systems by hop for Impact, a tree for Depends on, one line across boards for Route. Then a table of what is hit and how, with what is proven by code but not drawn marked, and what limits the result.
It ends with the next steps that fit: go one hop deeper, draw the result on a context board, or save a route as an insight. root lets the result cross apps; naming an app's board keeps it inside that app.
With your own MCP client
Any MCP client with a token can call the tools directly. Each takes workBoardSlug, the board whose tree to search, and from, the element; find_path adds to. Optional inputs set depth, the kinds to follow (through), the flags to add (include: work items, insights, boundary rules, owners) and limit. A route that doesn't exist comes back as found: false, not as an error.
When an agent records a route with create_insight, it can pass the result it came from as traceAsk. That insight then carries Ask this again, like one you saved yourself. create_insight needs a Read + write token. MCP Tools lists every tool and its scope.
FAQ
Why does an element I expected not appear?
Check that Scope includes its board, that Depth reaches it, and that Follow includes the kind of dependency that connects it. When reference links can't be resolved, the note under the list says so.
Is there a limit to how much a result covers?
The panel lists the nearest 200 elements and says when there are more. Chat and MCP list the nearest 60 unless the agent asks for more, up to 200. A scope holding more than 20,000 elements is refused; pick a board lower in the tree.
Do uncommitted changes count?
Yes. Elements and dependencies that are still proposals are included; the Depends on tree and a selected element mark them proposed.
Can I link to a result?
No. A canvas link can open the Dependencies panel, but not the element and type in it.
Is Dependencies on published boards?
No. It needs you to be signed in.
What's next
Chat
The same three types in your own words, as a dependency card under the reply.
Insights
Where a saved route lives, and how an insight is read on the canvas.
Aspects
The tables, endpoints, channels and pages that prove dependencies the diagram doesn't draw.
MCP Tools
The same three types as tools for an agent.
Last updated November 6, 2026