Core tools
Thirteen tools for an agent's everyday work in Cypress, and every other operation through discovery. What Cypress Operate shows an agent by default.
TL;DRAI summary
This chapter explains Cypress Operate's core tools, the 13 essential operations that help you manage your fleet efficiently. These tools simplify common tasks like finding vehicles, managing offers, and updating prices, acting as a streamlined interface for all operations.
Cypress Operate offers an organization about 400 tools, one per operation. That is every action of the app, but it is too many for an agent to choose from: the full list is about 640 KB, and models pick the right tool less reliably once they see more than a few dozen. The core set is the same Cypress in 13 tools, about 15 KB:
- eight named tools for the work an agent does every day, and
- discovery, three steps that reach every other operation: find it, read its inputs, run it.
It is what Cypress Operate shows an agent by default: connect to the server's address and the core tools are there.
https://app.getcypress.ai/api/mcp/operateAdd an area when a job needs that area's tools in view beside them, such as
?toolsets=core,planning. A gateway that brings its own tool search asks for
?toolsets=all: every per-operation tool instead, about 400.
/api/mcp/operate?toolsets=core names the default.
The core tools are not a second way of doing things. Each one calls the per-operation tools, which call the REST API as you, so the scopes, the rules, the dry runs and the decisions kept for a person are exactly theirs.
The eight named tools
| Tool | What it does | Runs |
|---|---|---|
find | Finds vehicles, listings, offers, closings, contacts, companies, deals, tasks, campaigns, plan vehicles or buy targets: by ids (up to 25), or by search, status, your own externalId and the type's own filters. A filter the type does not take is named with the ones it does | The type's list_ or get_ tool |
get_context | One record with what surrounds it, in one call: a vehicle with its listing readiness, offers, closings, deals and open tasks; an offer or a closing with its vehicle; a contact, company or deal with its deals, open tasks, notes and calls | The record's get_ tool and its neighbours' lists |
brief | What needs you today: Mission Control, the open tasks due within seven days, the closings waiting on your side | read_mission_control, list_tasks, list_closings |
launch | Puts one or up to 25 vehicles on the market, with their prices, channels and mode. The dry run says what each one lacks to go live | launch_vehicle, per vehicle |
reprice | Changes the buy-now price, the start bid or the floor of one or up to 25 vehicles | update_listing for a car with a listing made through the API; edit_vehicle for one without |
answer_offer | Accepts, counters, declines or withdraws an offer on your vehicle | accept_offer, counter_offer, decline_offer, withdraw_offer |
advance_closing | Moves a closing on: reminds whoever's turn it is, marks your own side's step, asks for a missing document, re-sends a signing link, reports an incident | remind_closing_step, mark_closing_step, request_closing_document, resend_signing_link, report_closing_incident |
log | Records work on a contact, a company, a deal or a vehicle: a note, a call or meeting, a task for a person or an agent, tags | The notes, interactions, tasks and tags tools |
A named tool never folds in a decision kept for a person. launch and
reprice answer NEEDS_PERSON for a floor from an agent's token, as the tools
they run do. advance_closing leaves out uploading documents, sending them for
signature and confirming that money arrived: those are a person's, in the app.
A tool given several vehicles answers one result each, DONE, NEEDS_PERSON
(with its decisionLink) or REFUSED (with the reason), and counts them. Given
one vehicle, it answers as the tool it ran. A retry with the same
idempotencyKey is safe: each vehicle gets its own key from it.
Everything else: discover, describe, run
Settings, members, planning, marketing, templates, automations and the rest are reached in three calls:
discoverwith what you want done, in words: "pause a car", "approve the quarter", "invite a member". It answers the operations that fit, among those your credential may use, each with one line and the executor that runs it. Narrow it withareaorrisk.describe_operationwith a name reads its inputs (JSON Schema), what its dry run checks and whether an agent may run it.run_read,run_writeorrun_destructiveruns it with its arguments. Calling one with the wrong executor answers which one to use:
{ "name": "pause_vehicle", "arguments": { "id": "clx…", "dryRun": true } }There are three executors so your client can ask you before anything
destructive by the tool's name alone: run_destructive removes, archives,
withdraws or cannot be taken back.
Discovery offers only what the credential may use: a scope it lacks hides the operation, and an organization's key is never offered Cypress staff tools. The answers are the operation's own, decision links included.
Workflows that use them
The Cypress Manual serves three skills that walk an agent through whole jobs with these tools: pricing and publishing cars, triaging offers, and chasing a closing. A client that supports skills loads the one a job needs.
Which to give an agent
| The agent | Ask for |
|---|---|
| Starting out, or doing a bit of everything | Nothing: the core tools are the default |
| Working one area all day (a closing coordinator, a planner) | core and that area: core,closings, core,planning |
| Running behind a gateway with its own tool search | all: every per-operation tool its scopes allow |
The scopes still decide what an agent may do: the core set shows less, it never allows more. See Keys and tokens.
MCP servers
Connect an AI agent to Cypress over the Model Context Protocol, to work in your organization or to read what Cypress knows.
Domain model
The records Cypress keeps, their ids, how they link, their states, what changes them and what they announce. The map an agent platform builds its own model from.