Give your agents a reliable answer to “who should act?”
Build agents that can resolve the right approver, owner, reviewer, group, role, or delegated contact without embedding organizational logic in prompts or application code.
Roster gives agents one participant-resolution surface across MCP and REST — backed by projects, directory data, participant metadata, memberships, and active delegations.
Your agent can know how without knowing who
Agent frameworks help you decide which tool to use, which action to perform, which data to retrieve, when to pause for human review, and how to continue after a decision.
But real workflows also ask questions the framework can’t answer on its own — who should approve this, who owns this customer, which team handles this exception, who is covering for the normal owner, who should receive the escalation.
Those answers change as employees move, group memberships shift, projects evolve, and responsibilities are delegated. Roster keeps that organizational logic outside the agent.
One governed tool for organizational routing
Use Roster’s resolve tool for open-ended responsibility questions, and exact tools when the agent needs to inspect specific resources. Every capability is authorized against the acting identity, credential scopes, project access, and resource rules.
Ask an open-ended question like “Who should approve the Atlas vendor renewal?” and receive an authorized, request-relevant participant.
Inspect the projects, participants, participant members, and labels the agent is authorized to see.
Check active delegations and coverage so the agent routes to the right person during absences.
Look up materialized users, groups, and memberships that back a participant selection.
Review previous resolutions attributed to the agent’s identity for debugging and audit.
Every capability is available through the MCP server and the REST API, using the same authorization model.
Build with the surface your agent already uses
Connect Claude, ChatGPT, Codex, Claude Code, or another compatible MCP client to Roster.
Call Roster directly from a custom agent, orchestration service, or backend.
Represent autonomous agents as dedicated Roster identities with their own credentials, roles, scopes, lifecycle, and audit attribution.
Run Roster with your approved OpenAI, Anthropic, Mistral, or compatible gateway-backed model configuration.
Example: agent-to-human approval
An agent prepares a vendor purchase and determines that human approval is required. Instead of reading a fixed approver from its prompt, it asks Roster:
What Roster evaluates
Every resolve request is answered against the acting identity’s authorized view.
- The project
- The relevant approval participant
- Participant labels and metadata
- Associated users or groups
- Materialized directory membership
- Active delegations
- The agent’s authorized visibility
Stop rebuilding organizational logic in every agent
Without a shared resolution layer, each agent ends up maintaining its own version of approver tables, group lookups, organizational rules, escalation mappings, absence handling, fallback contacts, and authorization assumptions.
Roster turns participant resolution into reusable infrastructure. One project model can serve multiple agents, multiple MCP clients, internal applications, workflow platforms, CI pipelines, and human users.
Built for production agent access
Give every agent its own Roster identity and API key rather than sharing a generic administrator credential.
Grant only the Resolve and lookup tools required for the agent’s job.
Handle no-match, out-of-scope, ambiguity, and provider failures as workflow states instead of letting the agent guess.
Connect model runs and Resolve requests to the acting agent identity.
Roster does not charge per AI agent, service account, API request, MCP call, or resolution.