Learn Shell

What sf-shenfeng/learn-shell MCP server does

The sf-shenfeng/learn-shell MCP server provides a persistent learning workspace for an AI agent and a learner. It stores courses, lessons, exercises, flashcards, live sessions, and teaching-related memory so a later session can continue from the existing state. A human learner uses the web application, while the teaching agent works through MCP; both surfaces read and write the same underlying data.

The system covers a broad teaching workflow. An agent can author course material and exercises, grade learner responses, create flashcards from lesson content, conduct real-time turn-based teaching, and read back learner context. The product also includes pages for courses, lessons, reviews, cards, a journal timeline, quizzes, mind maps, and document reading. Lesson completion is assigned to the learner rather than inferred solely from an agent's action.

How it works

MCP is the agent-facing protocol. The live registry exposes 50 tools at the time described by the repository, along with pair:// data resources, per-recipe recipe:// resources, and a manifest://capabilities resource generated from the running registries. Mutating operations return structured success or error envelopes. The documented failure categories include validation, not found, conflict, permission, and retryable errors.

Teaching memory is available through context read-back operations such as get_context and get_learner_brief. Returned information can include active learning contracts, recent lessons, pending work, live-session state, reminders, progress, and learner or agent identity references. This lets a new teaching session inspect the current relationship instead of reconstructing it from an earlier conversation.

The server also supports a composable skill and prompt layer. Its documented dimensions include domain, modality, intensity, tone, and pace, selected per learning contract. The repository describes English-first documentation and receipts, with MCP tool descriptions and the preparation report Chinese-first in v0.1.

Setup and configuration

The sf-shenfeng/learn-shell MCP server is self-hosted and uses Postgres with a web application. The repository points to SETUP.md for a from-zero process covering Docker Postgres, database migration, seeding, running the application, and connecting an AI over MCP.

By default, the server binds to 127.0.0.1, matching its single-machine trust model. It does not provide an account system, per-agent credentials, scoped tokens, or an audit trail beyond the documented MCP actor.

Limitations and notes

The sf-shenfeng/learn-shell MCP server is described as v1 and not authenticated. Its security model assumes one machine and one trust boundary, so it is not presented as a multi-user or public deployment. Authentication for multi-device use is on the roadmap rather than a current capability.

The repository describes a working system used in one real teaching relationship, not a general security-hardened service. MCP specification changes are being monitored, but support for the cited 2026-07-28 revision is not promised. Localization is incomplete in v0.1.