Roster for CrewAI

Give every crew a reliable way to determine who should act.

Connect CrewAI to Roster over MCP so agents can resolve current owners, approvers, reviewers, escalation contacts, and delegates from governed organizational context.

CrewAI coordinates the agents. Roster resolves the people around their work.

Connect Roster to CrewAI Read the Roster MCP documentation

Last verified 07/14/2026

Agent roles do not answer organizational ownership

CrewAI lets you define agents with roles, goals, tools, and tasks, then coordinate them through crews and flows.

Those agent roles describe what an agent does. They do not identify the current person or group accountable for a real business decision:

“Who approves the final proposal?”

“Who owns this customer account?”

“Who should review the security finding?”

“Who receives the escalation if the workflow fails?”

“Who is delegated to act this week?”

If those answers are embedded in task descriptions, prompts, or Python code, they drift as teams and responsibilities change. Roster gives crews a reusable participant-resolution capability backed by current project, directory, role, membership, and delegation context.

How Roster fits into a crew

Attach Roster MCP tools to a crew's agent, or call them from a bounded step in a flow, wherever participant resolution belongs.

CrewAI crew or flow
        ↓
Agent with Roster MCP tools
        ↓
Roster Streamable HTTP MCP server
        ↓
Project + participant model
        ↓
Directory users, groups, roles, memberships, and delegations
        ↓
Resolved owner, approver, reviewer, or escalation contact

What CrewAI agents can resolve

Approval assignment

Let an agent find the eligible approver before a flow creates or waits for a human approval.

“Who should approve this vendor exception?”

Customer and account ownership

Resolve the current owner using project and organizational context rather than a name copied into a task.

“Who owns the Northwind account in Canada?”

Specialist review

Select the right legal, security, finance, procurement, or technical reviewer when another agent identifies a need.

“Who should review the data-retention terms?”

Flow escalation

Route exceptions or failed guardrails to the correct escalation contact.

“Who receives this unresolved billing escalation?”

Crew-to-human handoff

Identify the accountable person or group when a crew completes its automated work and needs a human decision.

“Who should receive the completed implementation plan?”

Delegated responsibility

Resolve an active delegate when the normal owner or approver is unavailable.

“Who is covering for the procurement approver?”

Connect Roster to CrewAI

CrewAI documents MCP servers as tools for agents. Connect a CrewAI agent to Roster's remote MCP endpoint using the MCP integration method documented for the CrewAI version you deploy.

The connection requires:

The Roster Streamable HTTP MCP URL for your deployment.
An acting Roster identity with permission to access the required projects and participants.
OAuth or a Roster API key sent as a bearer token, according to the authentication flow used by your deployment.
Only the Roster tools required by the agent.

The acting identity requires mcp:resolve to call the resolve tool. Additional exact lookup tools require their documented mcp:* read scopes.

CrewAI has introduced more than one MCP integration style across releases. This page intentionally does not provide a version-specific Python or YAML snippet. Follow CrewAI's current MCP documentation for the client syntax, and use Roster's current MCP documentation for the endpoint and authentication values.

Resolve-only automation scope

For resolve-only crews, keep scopes narrow. Effective access remains bounded by the acting Roster identity's permissions and target-resource rules.

mcp:resolve
mcp:resolve-requests:read
mcp:projects:read
mcp:participants:read

Give one agent responsibility for participant resolution

In larger crews, use a dedicated routing agent or a clearly bounded task for organizational decisions. This keeps participant resolution consistent and prevents every agent from independently inferring ownership.

Recommended instructions for the routing agent:

- Call Roster for current ownership, approval, review, or
  escalation questions.
- Provide the business context needed for resolution.
- Include active delegations where applicable.
- Return no-result and authorization failures explicitly.
- Never reconstruct current assignees from memory or stale
  task content.

Example: add a legal reviewer to a crew

A proposal crew finishes research and drafting. A guardrail identifies that the proposed terms include a non-standard data-retention clause.

“Who should review the Northwind proposal's non-standard data-retention terms?”

The routing agent calls Roster:

resolve(
  query: "Who should review the Northwind proposal's non-standard data-retention terms?"
)

Roster evaluates the relevant project, legal-review participant, region, labels and metadata, active membership, and delegations.
The flow creates a review step for the resolved participant.
The crew resumes only after the human outcome is returned through the surrounding workflow.

The crew never guesses the reviewer, and the routing decision is auditable through Resolve Requests.

Roster complements CrewAI

CrewAI remains responsible for agent roles, tasks, tools, collaboration, memory, guardrails, crews, and flow execution.

Roster provides a governed organizational-context layer when those agents need to select a real person, group, role, or delegate. It does not replace crew design or workflow orchestration.

Security and governance

Separate acting identities

Use a distinct Roster identity for each deployed crew, service, or security boundary where attribution matters.

Least privilege

Limit scopes and project access to the context the crew needs.

Controlled tool exposure

Attach Roster tools only to the agents responsible for participant lookup or resolution.

Secret handling

Load API keys from the deployment environment or secret manager, not task YAML or source control.

Failure as an outcome

Treat no result, out-of-scope, and authorization failure as flow states rather than invitations to guess.

Reviewable decisions

Use Resolve Requests to inspect the recorded resolution status, actor, project, query, and result counts visible to the acting identity.

Frequently asked questions

Yes. Attach Roster MCP tools to an agent inside a crew or call them from a bounded step in a flow, depending on where participant resolution belongs.

Connect your crews to current organizational responsibility

Let CrewAI coordinate the work while Roster resolves the people, groups, roles, and delegates responsible for the human decisions around it.

Connect Roster to CrewAI Read the Roster MCP documentation Explore all integrations