Add Perplexity sample agent (Python)

[Yogeshp-MSFT]

Adds a Python sample agent using Perplexity AI (Agent API / Responses API) with the Microsoft Agent 365 SDK.

What's included

• Full agent implementation using OpenAI SDK pointed at Perplexity's /v1/responses endpoint
• MCP tool integration with Mail and Calendar servers
• Multi-turn tool-call loop with automatic argument enrichment and auto-finalize for incomplete workflows
• Retry with exponential backoff + jitter on transient MCP errors (502/503/504)
• MCP session caching across turns with automatic cache invalidation on failures
• Observability scopes (InvokeAgentScope + InferenceScope) and token caching
• Email and Word comment notification handlers
• Typing indicators, install/uninstall handling, per-turn client cleanup
• Comprehensive README with Quick Start, Playground testing, deployment (Azure/GCP), and troubleshooting

Files

File	Purpose
main.py	Entry point, observability config, shutdown handler
hosting.py	Message routing, typing indicators, notifications
agent.py	Perplexity orchestration, MCP tool discovery, system prompt
perplexity_client.py	Agent API wrapper, tool loop, enrichment, auto-finalize
mcp_tool_registration_service.py	MCP JSON-RPC client, schema sanitization, retry logic
agent_interface.py	Abstract base class
token_cache.py	Observability token cache
ToolingManifest.json	MCP server declarations (Mail, Calendar)
pyproject.toml	Dependencies
.env.template	Environment variable reference
README.md	Full documentation

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds a Python Perplexity-based sample agent using the Microsoft Agent 365 SDK, including MCP tool integration, observability, notifications, and end-to-end setup documentation.

Changes:

• Introduces a Perplexity Responses API client with a multi-turn tool-call loop, argument enrichment, and auto-finalize behavior.
• Adds MCP server discovery + lightweight JSON-RPC client with retry/backoff and session caching across turns.
• Provides runnable aiohttp hosting, observability token caching, and comprehensive documentation + environment/dependency setup.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
python/perplexity/sample-agent/token_cache.py	Adds a simple in-memory cache for observability exporter tokens.
python/perplexity/sample-agent/pyproject.toml	Defines dependencies and uv configuration for the sample agent.
python/perplexity/sample-agent/perplexity_client.py	Implements Perplexity Responses API wrapper and tool-call orchestration loop.
python/perplexity/sample-agent/mcp_tool_registration_service.py	Implements MCP server discovery, schema sanitization, caching, and retrying tool executor.
python/perplexity/sample-agent/main.py	Adds server entry point, auth middleware setup, health endpoints, and observability config.
python/perplexity/sample-agent/hosting.py	Adds message routing, typing indicators, token exchange/caching, and notification handlers.
python/perplexity/sample-agent/agent_interface.py	Defines the async agent interface used by the hosting layer.
python/perplexity/sample-agent/agent.py	Orchestrates Perplexity calls, MCP tool wiring, and observability scopes.
python/perplexity/sample-agent/ToolingManifest.json	Declares default MCP servers (Mail, Calendar).
python/perplexity/sample-agent/README.md	Adds Quick Start, Playground testing, deployment, and troubleshooting docs.
python/perplexity/sample-agent/.gitignore	Ignores local secrets, venv, and generated deployment artifacts.
python/perplexity/sample-agent/.env.template	Documents required environment variables for local and production use.

[Yogeshp-MSFT]

Hi @gwharris7 can you review this pr?

[Yogeshp-MSFT]

HI @biswapm can you review this pr?