Process and query email across providers via MCP agents
I have built an intelligent email product based on MCP and agents, which will be available initially as a ChatGPT app. Task: The author developed a provider-agnostic email intelligence product using MCP and agents, initially targeting a ChatGPT app interface. Development spanned six months using Cursor, Codex, and Claude Code, resulting in a cloud-hosted system supporting Gmail and Outlook. The architecture prioritizes privacy by processing emails transiently via open-source models without storage. The product is currently in private beta, awaiting platform approvals from Google, Microsoft, and OpenAI.; tools: Cursor, Codex, Claude Code. Human role: The original post does not say how a person stayed involved.; limitations: The product is restricted to private beta status, limiting user feedback and market validation. Reliance on open-source models for transient processing may introduce latency or quality variability compared to proprietary solutions. Integration with Gmail, Outlook, and OpenAI requires strict compliance with their respective API and privacy policies, which are currently blocking public release..
Tools
Published on Agent Work Atlas:
Overview
I have built an intelligent email product based on MCP and agents, which will be available initially as a ChatGPT app.
Workflow
The author developed a provider-agnostic email intelligence product using MCP and agents, initially targeting a ChatGPT app interface. Development spanned six months using Cursor, Codex, and Claude Code, resulting in a cloud-hosted system supporting Gmail and Outlook. The architecture prioritizes privacy by processing emails transiently via open-source models without storage. The product is currently in private beta, awaiting platform approvals from Google, Microsoft, and OpenAI.
Reported result
No measured business outcomes or user metrics are reported. The system is functional in a private beta environment but has not yet launched publicly due to pending approvals from major platforms. The author notes the approval process is tedious and is actively seeking a co-founder for 50/50 equity to continue development.
Human role
The original post does not say how a person stayed involved.
Limitations
The product is restricted to private beta status, limiting user feedback and market validation. Reliance on open-source models for transient processing may introduce latency or quality variability compared to proprietary solutions. Integration with Gmail, Outlook, and OpenAI requires strict compliance with their respective API and privacy policies, which are currently blocking public release.
Try it
Act as a technical lead for an email intelligence agent project. I need to prototype a privacy-focused email assistant using MCP and open-source models. Inputs: [Gmail/Outlook API credentials], [Open-source LLM endpoint], [MCP server configuration]. Steps: 1. Set up a local MCP server that connects to the email provider APIs. 2. Configure the agent to process emails transiently without storing content on disk. 3. Implement a query interface for the agent to summarize and extract action items from recent emails. 4. Run a dry run on a sample inbox with 5 emails. Output the JSON structure of the extracted intelligence. 5. Verify that no email content is persisted in logs or temporary files. Deliverables: A working MCP server script, a sample output JSON, and a privacy audit log. Success Criteria: The agent correctly summarizes 5 sample emails without storing raw content. Constraints: Do not send any emails. Do not use proprietary LLMs for the core processing logic. If API limits are hit, pause and report. Ask clarifying questions if the email provider API scope is unclear.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.