LFX 2026 Proposal: End-to-End Agentic Workflow

[lkingland]

Description:

Knative Functions ships an MCP (Model Context Protocol) server that exposes its toolchain to AI agents. This project builds on that foundation: deepening the MCP server to cover initial environment setup and CI/CD integration, and authoring a companion skill that walks agents through the full lifecycle of a Function. Together these enable end-to-end agentic usage of Serverless Functions; from initial scaffolding through deployment with CI/CD.

Expected Outcome:

• New MCP operations covering prerequisite checks with guidance.
• A companion skill that composes initialization, CI/CD setup, and deployment into a guided workflow.
• End-to-end demonstration.
• User-facing and agent-facing documentation.

Recommended Skills:

• Familiarity with the Go programming language (ideal) or Python (secondarily).
• Experience with AI/ML agents and interest in programmatic LLM integrations.
• Familiarity with Kubernetes, Serverless, GitOps, CI/CD systems a plus.
• Strong communication skills, with the ability to research and document clearly.

[Itx-Psycho0]

Hi @lkingland and @gauron99
I'm Anurag. I contributed to CNCF Knative in Term 1 2026 with merged PRs in the Knative Educational Game project.
I'm interested in this mentorship project. The idea of deepening the MCP server for end-to-end agentic usage of Serverless Functions is something I find really exciting, especially given my background in AI/ML agents and LLM integrations.
To get familiar with the codebase I've already studied the repo architecture, CONTRIBUTING.md and AGENTS.md, assigned myself to issue #3515 (Remote Quarkus build fails on s390x/ppc64le clusters) and am actively investigating the FSGroup permission issue in the pipeline code, and set up the full local dev environment including Go, Docker, kubectl and kind.
I'd love any guidance on where contributions would be most impactful toward the agentic workflow goal, particularly around the new MCP operations for prerequisite checks. Happy to start there if that's a good entry point.
Looking forward to contributing!

[RayyanSeliya]

Hey interested candidates I would suggest to read the existing mcp integration and understand it thoroughly u can refer https://github.com/knative/func/blob/main/docs%2Fmcp-integration%2Fintegration.md
And also the command related to mcp here https://github.com/knative/func/blob/main/cmd/mcp.go. And the mcp test code too here https://github.com/knative/func/blob/main/cmd/mcp_test.go

This will give u an idea of existing behaviour and also try to play with this command and write a detailed proposal of your approach towards the project !! Thx !

[ishanvaibhav]

Hi @lkingland and @gauron99,

I’m very interested in the LFX 2026 project around the end-to-end agentic workflow for Knative Functions.

I’ve started exploring the current MCP implementation, including:

• the MCP integration documentation
• cmd/mcp.go
• cmd/mcp_test.go

to better understand the existing agent interaction model and available toolchain operations.

What makes this project especially exciting to me is the combination of:

• AI/LLM agent integrations
• workflow orchestration
• Kubernetes/serverless tooling
• and developer experience automation through MCP

From reviewing the proposal, I understand the goal is not only exposing additional MCP operations, but also designing a cohesive guided workflow where agents can:

1. validate prerequisites,
2. scaffold functions,
3. configure CI/CD,
4. deploy,
5. and guide users through failures or missing dependencies.

I’m particularly interested in how the companion skill will compose these operations into a reliable agentic lifecycle instead of exposing disconnected tools individually.

A few implementation questions came to mind while exploring the current structure:

1. Do you envision the companion skill primarily as a deterministic orchestration layer over MCP operations, or should it also incorporate adaptive decision-making/recovery flows for partially configured environments?

2. For prerequisite validation, should MCP operations focus mainly on local environment inspection (Docker, kubectl, cluster access, registries, etc.), or also validate remote cluster/serverless readiness?

3. Are there plans to make the workflow provider-agnostic across different CI/CD systems, or would initial implementations target a smaller supported set first?

To prepare further, I’m planning to experiment with extending MCP-style workflows locally and studying how agent-driven execution paths can be made more resilient and observable.

I’d love to contribute incrementally and would appreciate any guidance on high-impact areas to start contributing around the MCP server or workflow composition pieces.

Best regards,
Ishan Vaibhav

[Nachiket-Roy]

Hi, I’m also very interested in this project. I’ve been going through the related codebase and MCP implementation to better understand the current architecture, existing gaps etc. I’ve also joined the #knative Slack workspace under the same name (Nachiket Roy) for communication and discussion around the project.
Thanks @RayyanSeliya for the guidance — it was genuinely very helpful in understanding the current system and shaping a proposal around the missing workflow.

Thanks again for the opportunity!

[gauron99]

Everyone interested please take a look at our CNCF slack channel to make your LFX applications on the their website 🙂