MCP Multi Tool Agent Platform
An agent platform built on LangGraph's StateGraph with a Model Context Protocol client for external tool access, async SQLite backed checkpointing for persistent multi session memory, and human in the loop interrupts before sensitive tool calls execute.
- Memory
- Persistent
- Approval
- HITL
- Tools
- Live
Selected engineering work by Muhammad Mubashir, showcased by Retrieva Labs. Figures and scope reflect the existing project notes. The evidence section explains what is available to review.
A useful outcome starts here.
An agent needed to use external tools and retain context across sessions while leaving sensitive actions under human control.
From input to useful output.
- 01Receive a task
- 02Load checkpointed state
- 03Select an MCP tool
- 04Request approval when needed
- 05Execute and preserve the result
The choices behind the system.
Explicit workflow state
LangGraph StateGraph makes the workflow state explicit rather than relying on a conversation loop alone.
Persistent checkpoints
An async SQLite checkpointer preserves conversation state across sessions and supports resuming an interrupted workflow.
Human control
An interrupt before a sensitive tool call makes approval an execution boundary rather than a request in a prompt.
Project notes, with context.
The published engineering note describes checkpointing, tool access and approval paths. Client source code and operational traces are not published in this portfolio.
Read the engineering note