Prompts and resources

Next to its tools, the server offers the two other MCP capabilities: prompts, the agency's recurring procedures written once and offered as commands, and resources, context documents a client loads once instead of calling tools for.

PromptsCopy Anchor Link

A prompt is a ready-made instruction that composes the existing tools. It grants nothing: the credential still decides what may be read or written, and a procedure the credential is not allowed to run fails exactly as the same calls typed by hand would.

List them with prompts/list; fill one with prompts/get, which returns the user message your client sends to the model:

{
  "jsonrpc": "2.0",
  "id": 3,
  "method": "prompts/get",
  "params": {
    "name": "weekly_client_update",
    "arguments": { "client": "hsbftmp3f5", "since": "2026-10-22" }
  }
}

A missing required argument is a JSON-RPC error (-32602) naming it. Arguments are strings.

Every prompt carries the same house rules for the model: work only from what the tools return and say what is missing rather than inventing it; read dashly://conventions before writing a status, moving a deal or assigning someone; cite the public id of everything it mentions; propose before writing anything back; answer in the language of the request.

weekly_client_updateCopy Anchor Link

What moved on one client over the week: requests opened and closed, time spent, what is left on the retainer, and what is late. Produces a short update an account manager can send.

prepare_meetingCopy Anchor Link

The state of one account before a call: open subjects, commitments made since last time, retainer left, open deals, and three questions worth asking. Produces a one-page brief.

estimate_requestCopy Anchor Link

What comparable work has actually cost the agency, from its own closed tickets and the time logged on them. Produces a range (shortest, median, longest) with the comparable items listed, or no number when fewer than three comparable items exist.

catch_up_on_late_workCopy Anchor Link

Everything past its due date and still open, grouped by client, with a proposed redistribution that looks at who already carries the late work.

pipeline_reviewCopy Anchor Link

Two lists of deals: those whose expected close date has passed, and those that have stopped moving, each with its value, owner and one next step. No forecast.

prepare_next_cycleCopy Anchor Link

A selection of backlog tickets for a project's next cycle, sized on the velocity its last closed cycles actually delivered. The selection is proposed; on a yes, only each selected ticket's cycle is set.

split_deal_into_partsCopy Anchor Link

Builds a complex project sold as several deals: one deal per milestone, recurring parts (maintenance, evolutions…) as recurring deals, follow-up deals on the follow-up pipeline. The parts are proposed as a table and written in a single action after a yes.

ai_visibility_setupCopy Anchor Link

Every AI-visibility setting of one client — brand, domain, competitors, engines, tracked questions, tracked keywords — read, reviewed against the latest results, and changed after a yes. It never claims the module is switched on when it is not (switching it on bills the client and is done in the dashboard), and runs a capture only when asked, since a capture spends the agency's allowance. See AI visibility from Claude.

ResourcesCopy Anchor Link

Four read-only JSON documents. resources/list answers without a credential; resources/read reads with the caller's credential, so each document only contains what that credential may see.

{ "jsonrpc": "2.0", "id": 4, "method": "resources/read", "params": { "uri": "dashly://conventions" } }

The tools cover the same ground on demand, which is what an agent needs after it creates a field or a stage. The resources exist so that an agent does not spend its first turns learning the shape of the product.

dashly://catalogCopy Anchor Link

Every resource of the API with, for each, its readable and writable fields, its filters, its sorts and its relations. Enough to write a correct call without guessing. The same content as the list_resources tool.

dashly://fieldsCopy Anchor Link

The built-in fields and the agency's own custom fields, per entity (client, deal, lead, ticket), with their key, type and options. This is what says how to fill customFields. The same content as the list_fields tool.

dashly://meCopy Anchor Link

Which agency the credential belongs to, what it is allowed to do (its scopes), and how much of the monthly token allowance is left. Read it first when a call is refused. The same content as the get_account tool.

dashly://conventionsCopy Anchor Link

The words this agency uses, which no schema can tell you:

The rules it spells out:

• A ticket's status must be one of workflowColumns[].id. Those keys are the agency's own.
• "Is this finished?" is isDone on the column, never a comparison with the literal done: an agency may have several terminal columns, and none need be called done.
• A deal's stageId must belong to the deal's own pipeline: match pipelineId on the stage before moving anything.
• Custom fields are not in this document; they are in dashly://fields.

Each section degrades on its own: if the credential may not read one of them (for example the members list), that section carries an unavailable entry with the refusal and the others are still returned.