Skip to main content
Unify MCP exposes 71 tools. Six manage the Agent lifecycle, four load identity or Unify data for context, and 61 provide direct access to Unify’s public APIs. Every tool call is scoped to the Unify user and tenant that authenticated the connection — a tool can only read or act on data that user can already see in Unify.

Agent tools

Use run_agent to start work, then use the other Agent tools to monitor, continue, cancel, and read runs.

run_agent

Starts Unify’s general-purpose Agent on a natural-language task. This is the same Agent behind Unify Chat, so run_agent supports anything Unify Chat does — research, building a list of companies or people, enrolling records in a Sequence, running a Play, writing updates back to your CRM, and similar work. Returns a run ID immediately; the Agent continues running in the background.
string
required
The natural-language task for the Agent to perform.
Starting a run requires the workspace to have chat funding/credits available, and is subject to per-user rate limiting. Both conditions surface as a tool error rather than a silent failure.
Example call
Example response

poll_agent

Checks the status of a run started by run_agent. Keep calling this while the run is PENDING.
string
required
The runId returned by run_agent.
Status values:
Example response (still running)

answer_question

Answers every question from a CLARIFICATION_NEEDED run and resumes it. Pass the questions back exactly as poll_agent returned them — the tool validates that the submitted answers line up with the pending questions before resuming the run.
string
required
The runId of the run awaiting clarification.
array
required
One answer for every question poll_agent returned, in the same order.
Example call

read_agent_results

Reads the final answer or error from a terminal run (READY or ERROR). Call poll_agent first — this tool errors if the run hasn’t reached a terminal status yet.
string
required
The runId of the run to read.
Example response

cancel_agent

Stops an in-flight run started by run_agent. It is safe to call after a run has already finished.
string
required
The runId returned by run_agent.

list_agent_runs

Lists Agent runs started through run_agent for the current user, newest first.
string
Exact run status to return. Omit to return all statuses.
number
default:50
Maximum runs to return per page. Cannot exceed 100.
string
The opaque nextCursor returned by the previous page.

Context and identity tools

get_me

Returns the authenticated Unify user’s profile and workspace identity. This tool does not require any input.

load_datatable

Loads a Unify DataTable the calling user owns, including its metadata, columns, and a cursor-paginated page of rows.
string
required
The ID of the DataTable to load.
number
default:100
Maximum rows to return per page.
string
The opaque nextCursor returned by a previous call, used to load the next page.
Only DataTables owned by the authenticated user are accessible — this tool can’t load another user’s tables even within the same tenant.

load_mailbox_voice_profile

Loads the writing voice profile for a specific connected mailbox — the same profile Unify uses to personalize outbound copy in that mailbox’s voice.
string
required
The ID of the connected mailbox to load the voice profile for.
The mailbox must be connected and owned by the authenticated user, and must already have an analyzed voice profile — otherwise the call errors.

load_general_context

Loads a specific version of the calling user’s general business context — the same information (business description, ICP, products, pricing, and so on) used to ground Unify’s Agents.
string
required
The ID of the user general context to load.
string
required
The specific version of that context to load.

Public API tools

These tools expose public API operations directly through MCP. Their input and output schemas are included in the MCP tool definitions, so your client can inspect the required fields before calling them.

Objects and records

Sequences and enrollments

Tasks

Events