Detailed Analysis
A Claude Desktop user has documented a practical challenge at the intersection of AI personalization and enterprise data boundaries: enabling a personal Claude instance to maintain contextual awareness of themes and priorities being handled by a separate, employer-provided Claude account, without merging the two environments or violating their required separation. The user's framing — borrowed from the television series *Severance* — captures the structural problem precisely: two cognitively isolated agents operating on behalf of the same individual, with no native mechanism for selectively sharing a summary layer between them. The proposed architecture involves scheduled automated tasks on both instances, where Work Claude generates a high-level brief and stores it to a shared intermediate location, and Personal Claude periodically ingests that brief to maintain ambient context.
The technical journey documented reveals meaningful gaps in the current MCP (Model Context Protocol) ecosystem. The user's first instinct — a Google Doc as a neutral, persistent, human-readable shared medium — was blocked by the Google Drive/Docs MCP's lack of write support, a limitation that reflects the still-maturing state of MCP server implementations for major cloud storage providers. Webhooks combined with n8n automation proved too complex for the use case, and the user ultimately landed on Notion, which does offer a functional MCP integration with read/write capability. However, Notion's presence as a single-purpose intermediary in an otherwise lean personal stack is correctly identified as architectural overhead — a tool adopted not for its value but as a workaround for the absence of better options.
The Obsidian pathway the user considers is technically coherent but runs into a fundamental constraint of that tool's design philosophy: Obsidian's vault is local-first and device-bound by default, making it unsuitable for cross-machine, fully automated data exchange without additional sync infrastructure (e.g., iCloud, Dropbox, or the Obsidian Sync paid service). This is a recurring friction point for users who are drawn to Obsidian's structured markdown approach and plugin ecosystem for AI-adjacent workflows but find its local-storage model incompatible with multi-device or multi-agent automation pipelines.
The broader pattern this post represents is increasingly common as AI assistant adoption scales within organizations: individuals operating at the boundary of professional and personal AI use are constructing informal data pipelines to compensate for the absence of sanctioned, privacy-respecting context-bridging tools. The "two Claudes" problem is a specific instance of a more general challenge — how does a user maintain coherent long-term context across isolated AI environments that are intentionally partitioned for compliance or privacy reasons? The solutions being improvised (Notion as a write-accessible neutral store, scheduled summarization tasks, MCP-based automation) are all workarounds for a capability gap that no major AI provider has addressed directly through native tooling.
Anthropic's Claude platform, through features like scheduled actions and the expanding MCP ecosystem, is enabling increasingly sophisticated personal automation, but this post illustrates that the connective tissue between enterprise and personal AI deployments remains underdeveloped. As more workers operate dual AI environments — one employer-provisioned, one personally maintained — demand will grow for lightweight, privacy-respecting context-bridging mechanisms that do not require adopting extraneous third-party services. The ideal solution the user is searching for, a writable shared document layer accessible to both instances via MCP, likely points toward either expanded Google Drive MCP write support or a purpose-built "AI context relay" abstraction that does not yet exist in any major platform's offering.
Read original article →