Layi Docs

MCP Integration

Expose installed Layi service-template actions as controlled tools for frontier agents and copilots.

Last updated: June 2026

Agent-readable tools for installed apps

MCP is the best option when another agent, copilot, or frontier model needs to discover what a buyer’s installed Layi service can do. Each installed app exposes deterministic tool metadata backed by the same service-template contract used by the dashboard, SDK, and API.

For direct backend calls, use API Integration. For Python helpers around the same surface, use the SDK.

Deterministic names
Tool names include service and action identity, so agent planners can call the right capability reliably.
Policy-filtered discovery
Agents only see tools enabled for the installed service, tenant, and MCP surface.
Same execution gates
MCP calls still pass through tenant scope, approval checks, compliance checks, and audit logging.

Routes

GET/services/{service_id}/mcp/tools

List the MCP tools exposed for one installed service.

POST/services/{service_id}/mcp/tools/{tool_name}/call

Call one tool with JSON arguments.

List tools

Start with tool discovery. Tool names are deterministic and normalized from the installed service and action key, for example layi_service__svc_123__talent_list_volume_tier.

curl -s   -H "X-API-Key: pk_xxx:sk_xxx"   https://agent.layi.ai/services/svc_123/mcp/tools

Call a tool

MCP calls accept JSON arguments. The runtime validates schema, tenant scope, action access class, approval state, and compliance metadata before returning a result.

curl -s   -H "X-API-Key: pk_xxx:sk_xxx"   -H "Content-Type: application/json"   -d '{"arguments":{"tier":"senior"}}'   https://agent.layi.ai/services/svc_123/mcp/tools/layi_service__svc_123__talent_list_volume_tier/call

Access behavior

  • Read tools by default

    Discovery, list, lookup, summarize, and inspect-style actions are the safest starting point for external agents.

  • Write/admin/export require allowlists

    Sensitive tools are hidden unless the installed template, tenant policy, and integration surface permit them.

  • Approvals still apply

    If an action requires approval, the MCP call must include approved execution context or the runtime blocks it.

Approval-gated MCP calls

When an exposed tool maps to a sensitive action, include approval context after a human or policy system has approved the operation.

curl -s   -H "X-API-Key: pk_xxx:sk_xxx"   -H "Content-Type: application/json"   -d '{    "arguments": {"pay_period":"2026-06"},    "approved_execution": true,    "approval_note": "Approved by payroll owner in ticket PAY-1024"  }'   https://agent.layi.ai/services/svc_hr/mcp/tools/layi_service__svc_hr__hr_run_payroll/call

Recommended agent posture

  • Use read-only first

    Start pilots with discovery and read tools before exposing state-changing operations.

  • Keep customer keys separate

    Use customer-scoped X-API-Key credentials for external tool access; never pass internal admin keys to frontier agents.

  • Log partner references

    Include a partner job or task reference when available so audit trails can be reconciled across systems.

Services & Apps

Install and operate the service templates before exposing them externally.