"Shell" matching MCP servers:

A Streamable HTTP MCP server that packages QMD search and document retrieval with bounded index-maintenance operations, without exposing arbitrary shell execution.

A local smart tool collection based on MCP that enables AI assistants to perform file management, document editing, shell execution, semantic retrieval, and text processing.

Local semantic search over Claude Code sessions and shell command history, exposed to Claude Code as an MCP tool. Everything is indexed into one vector space and runs entirely on your machine.

Local-first operator MCP server: recall across your own sessions, chats and files, an encrypted vault, SSH to your own hosts, and gated shell, browser and device tools. 130 tools over stdio or HTTP; everything runs on your machine — no hosted service, no account, no API key.

Indexes git history, shell commands with exit codes, project docs, and opt-in assistant transcripts into a local SQLite database, then serves ranked, token-budgeted context back over MCP. Uniquely links a failed shell command to the commit that fixed it, so agents can recall what was already tried and what worked.

An all-in-one MCP server for local LLMs that provides live web access, math, shell/code execution, filesystem access, memory, and RAG over local files using LM Studio embeddings.

Enables MCP clients such as Claude Desktop, Claude Code and Cursor to hand tasks and questions to a self-hosted agent that answers using its own local model, memory, document library and tools. It also lets those clients search long-term memory, save facts, and search, list or add documents, while shell commands and file writes are refused.

An MCP server over stdio that lets agents find code by meaning using local Ollama embeddings and trace callers or callees through real tree-sitter AST parsing rather than regex or keyword matching. It shares the same search and trace engine as the companion CLI, so MCP-native agents and shell-only agents get identical, reproducible results.

Exposes a read-only virtual filesystem over MCP so agents can explore documentation with real shell commands (ls, cat, grep, head, sed) answered from a vector store rather than a disk or sandbox. It also provides a hybrid dense-vector plus BM25 search tool for finding where to look first, with page access governed by group rules.

Enables agents to search a catalog of SKILL.md procedures, fetch individual skill instructions and supporting resources, and obtain advisory multi-step routes through read-only tools such as search_skills, route_request, get_skill, and get_skill_resource. It returns metadata and suggestions only, never executing skills, running shell commands, or granting authority.

Extends VS Code Copilot with MCP tools for persistent semantic memory, code indexing and RAG search, sandboxed Python/shell execution, web fetch/search, dynamic tool management via a web UI, and structured multi-step reasoning.