Rally API, MCP, and Agent

Three ways into the same infrastructure. Here is what each one does, who it is for, and how they work together.

Product Overview  ·  August 2026

Start here

One system. Three front doors.

Rally is one system: your participant CRM, your study templates, your consent records, your governance rules, your incentive rails, your audit trail. That system does not change based on how you reach it. What changes is who is doing the reaching.

API
For your systems.
MCP
For the AI tools your team already uses.
Agent
For the people who need a study and do not know your process.

The one-line version

What each one is.

API

A direct, programmatic way for your own systems to read and write data in Rally. Your developers decide exactly what gets called and when.

MCP

Connects Rally to the AI tools your team already uses, like Claude, so the AI can pull research data and take actions in Rally on your behalf, deciding which steps to take based on your instructions.

Agent

Rally's built-in Research Ops Agent that runs research work for you: setting up studies, managing recruiting, and handling day-to-day research tasks through conversation.

A useful way to tell them apart: the API runs the same steps every time because your engineers wrote those steps down. The MCP runs the steps your AI tool decides on, in the moment, based on what you asked for. The Agent already knows the steps for building and recruiting a study, so you only describe the audience and the question.

Side by side

How they compare.

API
MCP
Agent
Who it's for Engineers and data teams, building to a spec that Research Architects define Product builders and researchers working inside their own AI tools. Research Architects wire it into shared team workflows Product builders and researchers who need a study without learning your process. Research Architects set the rails everyone runs inside
Where it runs Your infrastructure, calling Rally Claude, ChatGPT, Cursor, Figma, Notion, Linear, or your own internal tooling The Rally Agent Command Center inside the Rally platform, or through the Rally Slackbot
How it behaves Deterministic. The same request returns the same result every time Composable. You direct it, and it chooses which Rally tool calls to make at each step Guided. It builds the study from your approved templates and walks you through approvals
Setup Generate an API key in workspace settings, then build the integration Connect in under 5 minutes from the help center instructions Ask your CSM to enable your workspace for the beta
Strongest at Panel sync, compliance exports, event-driven integration, reporting pipelines Recruitment and study management as one step inside a larger workflow you designed End-to-end study execution for someone who has never built a study

How they fit together

Not tiers. Not alternatives.

They sit on the same foundation and do different jobs.

API
Keeps the foundation accurate

Your CRM or data warehouse is the source of truth for who your users are. The API mirrors that into Rally on a schedule, so the panel you recruit from reflects reality rather than a CSV someone exported in March.

MCP
Puts it to work where your team already is

Exposes recruitment, study management, screening, scheduling, and incentive tool calls to any MCP-compatible AI tool, with your governance applied on every call.

Agent
Rally's out-of-the-box research workflow

Lives in the Rally Agent Command Center, a focused workspace built for product builders, and in Slack through the Rally Slackbot. The fastest path for teams who want an agent rather than a set of tools to build with.

Foundation
Rally Research

Participant CRM, governance rules, consent management, study templates, permissions, incentive rails, audit trail. Every other surface reaches this same system.

The pattern we see at scale: the API keeps the panel current, and the MCP and Agent recruit against it.


Why ours is different

Depth over breadth.

Many MCP servers are built the same way: wrap an existing API with tool calls and ship as many as possible. The result is a long list of capabilities that are technically functional but shallow in practice. Agents can call them, but they cannot do much with them.

The common approach
Built around "what can we expose?"

Tool calls mirror API endpoints one for one. A recruitment call returns a list of people. Whether those people are in cooldown, whether consent is current, and whether this person is allowed to contact them are all questions left for the researcher to answer, or left unanswered.

Rally's approach
Built around "what do agents actually need to kick off research?"

When you build for agents rather than for API coverage, you invest differently. Individual tool calls get deeper. They carry more context, enforce more rules, and handle more of the workflow autonomously so the researcher does not have to fill in the gaps.

A tool call that recruits participants does not just pull a list. It checks eligibility, applies cooldown windows, validates consent status, and enforces the criteria your Research Ops team configured in Rally before a single invitation goes out.

That is where our investment goes. Not into the number of tool calls, but into how much each one can actually do. The result is a focused set of tools built specifically around recruitment and governance, the two areas where research programs most often break down and where the consequences of getting it wrong are highest.

Read the full tool call reference at rallyuxr.com/post/rally-mcp-tool-calls-explained

Governance

Governance built in, not added on.

Read this section closely if you are the person who signs off on research access. These controls are properties of Rally Research, not features of any one surface, so they apply the same way whether the request came from a script, an AI tool, or the Agent.

Eligibility is applied before contact

Cooldown windows, opt-out and consent status, and the criteria your Research Ops team configured are all checked before a single invitation goes out. A person in cooldown is not surfaced as an option to skip past. They are not reachable.

Templates carry your standards

Approved study templates bring your consent language, incentive rules, and interview structure with them. Someone who has never run a study inherits your standards without needing to know they exist.

Permissions are inherited, not re-declared

Role-based access carries over from Rally. If a product manager cannot launch a paid external panel or set an incentive amount in Rally, that option is not available to them through the MCP or the Agent either.

Outreach goes out in waves

Recruitment sends in batches rather than contacting everyone who matches. A study that needs 8 participants does not invite 300 people to get them.

The first outreach round waits for a human

Proposed sends and a study's first round hold for approval in Rally. If nobody approves, nothing sends.

Money requires explicit confirmation

Starting paid external-panel recruitment requires confirming on Rally's cost sheet. Reviewing a quote does not reserve funds or notify the panel provider. The Agent can propose spend. It cannot commit it.

Everything leaves a trail

Each participant record carries a governance activity feed: when their contact status changed, who changed it, and why. When product builders run their own research, that history lives inside Rally for legal and compliance review.

The agent does not draw the line. You do.

What each one covers

The capability map.

API
  • Panel syncImport people, manage person properties and property groups, search people with filters or natural language, delete people, and read or update opt-out status in both directions.
  • Event-driven integrationWebhooks fire on participant status changed, person opted out, and import finished. If you would rather not run webhook infrastructure, poll the cursor-paginated status-event endpoints for participants and studies over a moving time window.
  • Compliance and reportingExport consent submissions as PDFs for one person or an entire study, search consent and signup form submissions, pull screener and survey responses with question-level answers, and retrieve observer rooms and interview transcripts.
  • Scripted study setupCreate draft studies, build and replace screener questions, publish a draft study, and change study status. Interviews, teams, segments, populations, and study templates are readable for reporting and provisioning.
  • Incentive operations at volumeSend gift card incentives to study participants, log custom incentives fulfilled outside Rally, resend a redemption email, and cancel an unclaimed incentive.
MCP
  • Feasibility before commitmentcheck_recruiting_feasibility tells you whether an audience is reachable from your panel and recommends a route: recruit from the panel, import a list, or use a paid external panel. Teams use this to choose a recruitment approach before building anything.
  • Study construction from a descriptionBuild or edit the study, write the screener or survey, attach a consent form, and configure incentives and the scheduler. Fields lock progressively as recruiting advances, and the assistant tells you when a change is no longer possible.
  • The full recruitment loopBuild the recruiting plan and send cadence, preview the outreach copy a participant will actually see, launch, pause and resume campaigns, configure external panel targeting from the provider's real option values, and review the itemized cost quote before launch.
  • Live schedulingSearch open availability, book, reschedule, change host, add guests or silent observers, mark attendance, and generate a prep brief for an upcoming interview.
  • People and past researchSearch the workspace people pool, manage populations and imports, pull a participant's full profile and notes, retrieve transcripts, and read screener and survey responses.
Agent
  • End-to-end study executionFrom one description of the audience and the question, it builds the study from an approved template, recruits from your own panel or a third-party panel, handles outreach, manages scheduling and quotas, and sends incentives with automated reminders.
  • Study setup for people who do not know your conventionsA product manager describes their project scope in plain language. Your naming conventions, consent requirements, and incentive rules are already in the template they inherit.
  • Research status and session changes from SlackAsk about upcoming interviews, change a host, or add an observer without opening Rally.
  • A panel that improves with every studyScreener answers map onto participant profile properties, so the next person searching for that attribute finds those participants. Screener data stops dying with the study that collected it.
  • A command center built for buildersA focused workspace to prompt the agent, see upcoming calls, and track sessions across the team.

One boundary worth stating plainly: recruitment orchestration, outreach campaigns, external panel workflows, and live scheduling are MCP and Agent capabilities. They are not in the API. If the question is whether recruitment can be automated through the API alone, the answer is no.


Customer stories

How teams are using them.

CustomerWhat they're doingSurface
Ramp Hands off an audience and the agent runs recruitment through to incentives, invoked from Slack or Claude with progressive approval at each step. They describe what they want as a research ops secretary inside their community of agents. Agent + MCP
Figma Runs full end-to-end studies inside their own Claude instance, using their own tooling to source users and Rally MCP to import and launch. MCP
Procore PMs and designers describe project scope and the agent builds the study. Ops builds reporting dashboards on top. Agent
Accelerant Queries their product analytics MCP to find an audience that never lived in Rally, then hands it to Rally Agent to build and launch. Runs a weekly automated research digest into Slack. MCP-led
Twilio Screener generation on tenure and usage questions, plus automated incentive reminders replacing manual ops work. Agent
Perplexity Recruits through the MCP with a browser fallback, pushing hard on quota and scheduling behavior. MCP
Blackbaud Checks how many people are likely to qualify before a study kicks off, so teams choose their recruitment approach with real numbers rather than assumptions. MCP

FAQs

Questions we get asked.

We already have the API. What does the MCP give us that we cannot build ourselves?

The API covers people, studies, screeners, incentives, and reporting. It does not cover recruitment orchestration, outreach campaigns, external panel workflows, or live scheduling. To build agentic recruitment on the API, you would need to construct those yourself, along with cooldown enforcement, consent validation, eligibility checks, template locking, approval gates, and batched sending. The MCP is that work, already built and maintained.

Is the MCP just your API with a chat interface on top?

No. Most research MCPs are built by wrapping an existing API and shipping as many tool calls as possible, which produces a long list of shallow capabilities. Rally MCP tool calls do more per call. check_recruiting_feasibility returns a routing recommendation, not a row count. A recruitment call checks eligibility, applies cooldown windows, validates consent, and enforces your Research Ops criteria before any invitation goes out. That logic is not in the API surface.

If we adopt the Agent, do we still need the MCP?

They serve different needs and most teams end up with both. The Agent covers the standard shape of a study for someone who does not want to design a workflow, and it runs in the Rally Command Center and Slack. The MCP is what you use when the audience lives in another system, when recruitment is one step in a chain you built, or when your team works in Claude or Cursor and will not switch surfaces to run research.

What stops a PM from over-contacting our customers or overspending?

Several controls, and they stack. Eligibility rules make people in cooldown unreachable rather than skippable. Outreach sends in waves and holds the first round for approval in Rally. Paid external panel spend requires confirmation on Rally's cost sheet.

Underneath all of that, you decide what a product builder can reach in the first place. Custom roles and permissions control which areas of Rally each person can access and which actions they can take, including whether they can send incentives or use a paid external panel. You can also lock individual areas of a study template so they cannot be edited, such as the incentive amount and the consent form. A builder working from a locked template inherits those decisions and cannot change them, whether they are in Rally, the MCP, or the Agent.

How do we prove this to our compliance team?

Every participant record carries a governance activity feed showing contact status changes, who made them, and why. Research run through the Agent or MCP produces the same trail as research run by hand in Rally, because it is the same system. Rally is SOC 2 Type II audited annually, supports GDPR and CCPA requirements, enables HIPAA compliance through zero-retention LLM APIs, and holds contractual agreements with every AI subprocessor prohibiting use of customer data for model training.

Which surface is most secure?

Same infrastructure, same permission model, same audit trail across all three. The MCP and Agent act as the person using them and inherit that person's role. The API authenticates with a workspace-level key, so scope who holds it deliberately. Generating an API key requires a Developer or Admin role.

Can our team use their own AI tool, or do they have to use yours?

Their own. Rally MCP works with Claude, ChatGPT, Cursor, and any MCP-compatible platform, and connects with research tools like Maze, Ballpark, Outset, and Listen Labs so participants can be managed across study types without switching context.

How long does this take to stand up?

MCP connection takes under five minutes with a Rally workspace. Agent beta access takes a request to your CSM. The API takes as long as your integration work requires, which for a scheduled panel sync is typically a short project.

Does using AI tools with Rally mean our participant data trains a model?

No. Rally holds contractual agreements with every AI subprocessor prohibiting the use of customer data to train models. Your participants, studies, and findings stay yours.

Already a customer

Connect Rally MCP in under five minutes from the help center, or ask your CSM to enable Rally Agent beta on your workspace.

Evaluating Rally

Book a demo at rallyuxr.com/demo to see governance running across a live recruitment workflow.

Read more

Rally MCP  ·  Rally Agent  ·  Developer docs