mail-shadow-mcp: local read-only MCP server for secure email access
mail-shadow-mcp, from Dryas, is an MCP server that provides AI agents with controlled, read-only access to a user's mailbox by creating a local mirror for model consumption. It lets models query and extract conversation content and attachments through a limited API, without granting send or delete privileges. The design emphasizes a single-binary deployment and local processing, aimed at developers and AI power users who need structured email access for summarization, localization, and document-management workflows.
What tasks can you actually use it for?
The server supports AI-driven summarization, localization, bulk organization, and document extraction, enabling agents to process inbox archives at scale. Its design permits models to assign ratings, group threads, and extract structured entities from messages, making it useful for preparing datasets or producing batch summaries. Developers can plug agent workflows that need historical context, for example converting multilingual threads for localization pipelines or extracting contract fragments for downstream processing.
How fast and reliable is search and extraction?
Search relies on a local full-text index to return queries quickly on large archives, with documented performance at 100,000-plus messages. Attachment handling exposes raw payloads for downstream model processing, so extraction speed depends on local I/O and model workload rather than repeated remote IMAP requests. Retrieval latency is driven by disk and agent processing, which reduces external query overhead for bulk operations.
Does it require technical setup and how does it handle data?
The server integrates with MCP hosts and requires either building from source with a Go runtime or running a precompiled binary on a compatible OS, so deployment assumes developer familiarity. It stores a local shadow copy that agents query and permits soft-deleting items only from that local store, not the live mailbox. The API surface is intentionally minimal and auditable, letting teams inspect exactly what models can read before granting access.
Best suited to developers who manage local MCP deployments
mail-shadow-mcp is a practical option for developers and AI power users who need model-readable archives of email history integrated into MCP workflows. It suits teams comfortable with local deployment and agent integration; AI-generated summaries should be verified by a human before use. Users needing a point-and-click mailbox editor or direct send/edit capability will need additional tooling around the server.
