sally/agents_

Sally speaks MCP. Your agent already does.

Connect Claude or any MCP-capable agent to Sally and it gets real tools across projects, tasks, CRM, marketing, timesheets, and memory — over a hundred of them — under the exact permissions you grant.

mcp/session
> initialize
← session ready
> tools/list
← workspace.* project.* task.* crm.* marketing.* memory.* opportunity.* timesheet.*
> tools/call task.resume_context
← latest handoff · evidence · blockers · recommended next action
> tools/call crm.follow_up.add
← follow-up created · due Friday
> tools/call marketing.message.propose_send
← PENDING approval · visible in Control Center

Getting connected

From account to working agent in three steps.

1. Create access

Mint an MCP key in your Sally account, or let your agent connect through OAuth with scoped access. Keys can be restricted by workspace and revoked at any time.

2. Point your client

Add your Sally MCP endpoint to Claude or any MCP-capable client. Your agent lists the available tools and sees exactly what it is allowed to do.

3. Let it work

Your agent reads project context, creates and updates tasks, logs CRM activity, drafts marketing messages, writes memory — and raises approvals instead of taking risks.

The tool surface

Not a chat integration. The whole product, as tools.

Every product area in Sally is exposed as MCP tools. If a human can do it in the UI, an agent with the right permissions can do it over MCP.

Projects & tasks

project.list · project.runtime_context · task.create · task.move · task.handoff.create · comment.add

Full board control: statuses, priorities, labels, checklists, comments, dependencies, reordering, and structured handoffs.

CRM

crm.person.create · crm.deal.update · crm.activity.add · crm.follow_up.add

Organizations, people, deals, activities, and follow-up reminders — so customer context stays fresh without manual data entry.

Marketing

marketing.message.create · marketing.campaign.create · marketing.audience.sync · marketing.message.propose_send

Draft newsletters in the visual builder format, assemble campaign graphs, manage audiences — and propose sends that always wait for approval.

Memory & opportunities

memory.create · memory.search · memory.intake · opportunity.list · opportunity.recommend

Write durable facts, decisions, lessons, and brand voice. Turn meeting notes into structured records. Ask Sally what is worth doing next.

Timesheets & clients

timesheet.add · timesheet.report · client.create · client.update

Log time, pull workspace-wide reports, and keep client records current — useful for agencies billing real work.

Workspaces & structure

workspace.list · project.create · project.status.create · project.member.add

Set up and reshape structure where permissions allow: projects, statuses, members, and workspace context.

Agent-ready context

Agents start informed and hand off cleanly.

The most expensive part of agent work is re-establishing context. Sally makes context and handoffs first-class tool calls.

project.runtime_context

One call returns what an agent needs before touching a project: summary, statuses, open and blocked tasks, relevant memory, recent activity, blockers, approvals, trust policies, and the recommended workflow.

task.resume_context

Pick up any task cold: the latest handoff, evidence, approvals, blockers, decisions, recent comments, related memory, and a recommended next action.

task.handoff.create

When an agent finishes or gets stuck, it files a structured handoff: a concise summary comment, an optional blocker or approval request, a status move, recorded evidence, and a decision brief for Control Center.

Control

Autonomy you can dial, boundaries you can trust.

Proposals before actions

Agents can propose actions instead of executing them. You review, approve, or deny — and Sally records the outcome either way.

Trust policies per agent

Allowlist safe actions for auto-execution — creating tasks, adding comments, logging CRM activity — while everything else waits for a human.

Typed approval gates

Publishing, credentials, payment data, customer data, client decisions, live deploys, and destructive actions each have an explicit approval type. Mass sends are proposals by design.

Typed blockers

Missing access, failing tests, ambiguous requirements, pending client decisions — agents raise structured blockers instead of improvising.

Everything on the record

Agent connections, runs, and actions are visible, revocable, and written to the operational event ledger and audit log.

Real permission checks

Every tool call passes the same permission resolver as the web app. An agent can never do more than the human who authorized it.

next/action

Connect your agent this week.

Start a trial, create an MCP key, and give your agent its first real job — with every risky action gated behind your approval.

7 days free. No credit card required.