CSC510 Lecture 22

Bedrock AgentCore
(20261019)

Transcript

1. CSC 5100 Modern Software Engineering Lecture 22. Amazon Bedrock AgentCore Dr. Javier Gonzalez-Sanchez javiergs@calpoly.edu www.javiergs.phd ffi o ce: 14 -227
2. Previously ...
3. Assignment 04 ✅
4. Next :: Homework
5. Roadmap
6. Strategies 1 2
7. https://github.com /csc-sandbox/5100
8. Roadmap 2
9. Your Goal and Your Homework
10. Claude Desktop
11. Claude Desktop
12. Connectors ✅ 2
13. Our Lambda endpoints — 1 of 2 • TemperatureLambda → /5100/temperature • DominantA ectLambda → /5100/dominant-a ect • RobotDistanceLambda → /5100/robot-distance • RobotReachabilityLambda → /5100/robot-reachability • LidarClosestLambda → /5100/lidar-closest • LidarSafetyLambda → /5100/lidar-safety ff ff • DataFreshnessLambda → /5100/data-freshness
14. Our Lambda endpoints — 2 of 2 • CoordinateNormalizerLambda → /5100/coordinate-normalizer • EventRateLambda → /5100/event-rate • GazeClusterLambda → /5100/gaze-clusters • EncryptMessageLambda → /5100/encrypt • NearestSampleLambda → /5100/nearest-sample • MovingAverageLambda → /5100/moving-average • A04 target: GET + POST for every assigned Lambda.
15. Amazon Bedrock AgentCore Gateway
16. Step 01 We need a replacement for our MCP server... 17
17. Amazon Bedrock AgentCore | Gateway • It m kes existing cloud services ccessible to AI ssist nts without modifying their underlying implement tions • It repl ces our custom J v MCP server from Lecture 20.. • Exposes REST APIs nd AWS services s discover ble MCP tools. • Am zon Cognito nd OAuth 2.0 to v lid te ccess tokens. a a a a a a a a a a a a a a a a a a a a a a a • Cl ude Desktop ➡ AgentCore G tew y (MCP) ➡ API G tew y (REST) ➡ AWS L mbd
18. Amazon Bedrock AgentCore | Gateway 1
19. csc5100-mcp-gateway https://csc5100-mcp-gatewaykeijwqwkag.gateway.bedrock-agentcore.uswest-2.amazonaws.com/mcp 20
20. Step 02 The server will connect to our AWS lambdas 21
21. • A t rget connects AgentCore G tew y to n existing cloud service, m king its oper tions v il ble s MCP tools. • A single G tew y c n h ve multiple t rgets, e ch exposing di erent services. • T rgets describe existing REST endpoints nd becomes discover ble MCP tool. • An OpenAPI contr ct describes the intended service independently of its implement tion: oper tion n me, description, nd input p r meters. a a a a a a a ff a a a a a a a a a a a a a a a a a a a a a a a a a a • Cl ude Desktop ➡ AgentCore G tew y ➡ T rget ➡ API G tew y (REST) ➡ AWS L mbd a a Amazon Bedrock AgentCore | Targets
22. Amazon Bedrock AgentCore | Gateway 2
24. Claude Desktop: remote connector
25. csc5100-mcp-gateway OAuth Client ID: ck2e940r9lmlidrp36bqv7j4e 26
26. Amazon Bedrock AgentCore | Gateway 3
27. Connectors
29. The entire conversation USER “Convert 86°F to Celsius” Discover menu Select tool MCP tools/call(value=86, from=F, to=C) GATEWAY HTTP GET → TemperatureLambda LAMBDA {"result":30.0} “86°F is 30°C.”
30. Takeaway Claude selects a tool from its description and schema. Build: Java publishes the menu. or Buy: AgentCore publishes the menu. The "kitchen" stays the same: TemperatureLambda.
31. Homework | Due Friday
32. Questions
33. CSC 5100 Modern Software Engineering Javier Gonzalez-Sanchez, Ph.D. javiergs@calpoly.edu Fall 2026 Copyright Notice. These slides are provided exclusively for educational use in CSC 5100 within the California State University system. Redistribution or use for any other purpose is prohibited without prior written permission.