Streaming Tool Execution Progress (to user) for Long-Running Tool Jobs (Factual/Storm Agents)

[vishnudev-k]

Feature Type

New functionality

Problem Statement

We need a way to show the progress of long-running jobs to the user.
When we are running tasks over FastAPI/MCP. We need to implement a polling mechanism to update the graph with current status messages without a heavy overhead to the main thread.

Proposed Solution

Langgraph can stream the tool status with the help of stream_writer
But this flow is suitable for tools that run within the current agent.

from langgraph.config import get_stream_writer
writer = get_stream_writer()
writer(f"Progress : {prog_percent}%")

MCP servers offer a way to provide a status update using context.

await ctx.info(f"Progress : {prog_percent}%")

But this implementation doesn't support langgraph tools using MultiServerMCPClient (Still checking)

Alternative Solutions

We can also think about using message queues, which can also handle the server overload.

User Benefits

No response

Implementation Ideas

No response

Contribution

• I'm willing to submit a PR for this feature
• I'm willing to test this feature
• I'm willing to help document this feature

Additional Context

No response

[vishnudev-k]

https://github.com/ibm-messaging/mq-agentic-ai
Here is an example of using message queues with pubs and subs to solve the problem of long-running tasks.
I am trying to implement it locally as a proof of concept.

From the Readme >

The demo explores how IBM MQ can support asynchronous, agent-based architectures by addressing limitations in existing agent protocols like ACP and A2A, which use JSON-RPC over HTTP or Server-Sent Events. These protocols struggle with complex, distributed systems due to limited support for retries, persistence, queuing, durability, and resilience under high load. IBM MQ offers reliable state updates, durable asynchronous messaging, security, and transactional integrity, complementing existing protocols. The aim is to gather feedback and engage organizations, leveraging IBM MQ’s proven enterprise-grade messaging features for mission-critical agent systems, rather than replacing current approaches.

[vishnudev-k]

One other approach came across is using web-sockets for the communication between the agent and the tool. This will enable continuous connection and the channel can stream the progress and final message without the need of polling and re-connection.