DotsSoftwareInfra & DevOpsAI score · 62/100AI-assisted edit · no obvious prompt-injection riskX

Using dots conversational layer to orchestrate custom MCPs for cloud infra and coding agents

Voice or chat intent → Dot structures into task → MCP reads/writes state and dispatches cloud workers → Collects PR/CI/review evidence → Human confirms critical decisions

ExperiencedStepsWorkflowTip
Who
Engineers or tech leads managing cloud infrastructure and coding workflows
You get
Unified natural-language control of Cloudflare, AWS, Claude Code with verifiable execution evidence chain
You need
dots · Cloudflare · AWS · custom task board

Constraints

Requires: OpenAI dots access; Cloudflare and AWS accounts with API credentials; Claude Code cloud agent environment; self-built MCP server (connecting above services); self-built task board system (maintaining shared state). Specific implementation details not disclosed in source.

What they did

BroadsideCode uses OpenAI dots as a conversational layer, connecting via custom MCP servers to Cloudflare, AWS infrastructure, and Claude Code cloud coding agents, together with a custom-built task board system, forming a complete workflow from intent input through cloud execution to evidence verification.

How it went

The author states the system is 'not perfect' but is heavily used ('I use the F out of it'). The task board serves as the shared state center, cloud agents perform the actual work, and the conversational layer connects intent, execution, and verified results. The author wants OpenAI to continue improving dots for coding and engineering work. (author_report)

Worth knowing

1. Source does not provide specific implementation code or architecture details for the MCP server and task board. 2. Author mentioned 'Evidence below' but no specific evidence was found in the captured snapshot, possibly due to incomplete reply thread capture. 3. No cost, latency, or scale data provided. 4. No concrete failure handling or rollback case demonstrated. 5. This site has not tested any components; description is entirely based on author self-report.

Steps

  1. Describe work intent by voice or chat

    Describe the work to dots. Dot automatically turns intent into a structured task with: scope, acceptance criteria, required skills, agent roles, execution limits, and rollback instructions.

  2. Dispatch cloud agents via MCP and track state

    Through MCP, Dot reads and updates task state on the task board, dispatches work to cloud workers (Cloudflare, AWS, Claude Code), and tracks handoff from queued → running → review. Version-checked updates prevent agents from blindly overwriting each other's changes.

  3. Collect execution evidence and cross-verify

    After execution, collect multiple evidence types: run references, pull requests, CI results, independent reviews, and downloadable artifacts. The orchestrator checks that these records agree. Note: a dispatched job is not a completed task; green CI is not proof the requested artifact is correct.

  4. Human-in-the-loop for critical decisions

    Stay in the loop for decisions, spending, destructive actions, and production approval. The AI translates system state into clear next steps so you can run the workflow without sitting in front of every tool. The task board holds shared state; cloud agents do the work; the conversational layer connects intent, execution, and verified results.

Prompt you can copy

Reference prompt · Editorial suggestion。Confirm the destination, input variables and output format before use.

This block is the prompt. Ads are never placed here.

No prompt needed — just follow the steps.

This case describes a product usage pattern with no copyable prompt. Actual implementation requires building your own MCP server and task board system.

Source

用 Dot 经自建 MCP 编排 Cloudflare/AWS/Claude Code 云端 agent，产出 PR/CI 证据链 · BroadsideCode · X · 2026-10-24T06:31:35.000Z

Comments & feedback

Tried this method? Share results, problems or improvements. Comments appear after moderation. Please avoid sensitive information.

Loading…

    Related