MCP servers that ship as code — a repository or a package you install and run locally (usually over stdio). They have no hosted endpoint, so we cannot probe or grade them: instead each one shows its repository, install command, license, stars and when it was last updated.
Collected from GitHub repositories and npm, PyPI, Cargo, Go, Maven, NuGet and RubyGems packages tagged as MCP servers, the official MCP Registry, Docker's MCP catalog, ModelScope, Baidu MCP World, Cline and other public directories; one entry per repository. A server that also offers a hosted endpoint is listed under Remote instead. These entries are not counted in our reachability, verification or readiness statistics.
-
A secure Python MCP server for Proton Bridge with a deny-by-default security model, opt-in SMTP capability, and email body injection fencing.
★ 5 -
terminusgithub.com/interceptai-tech/terminusupdated this year
The SQL firewall for AI agents: a default-deny sidecar / MCP enforcement point that governs what an AI agent runs against your database. Open core, AGPL-3.0.
AGPL-3.0 -
Local credential broker for AI agents. Keep API keys and SSH keys in an encrypted vault on your own machine; give each agent its own scoped, expiring, revocable grants; let it use a credential over MCP or ssh-agent without ever seeing it. Deny by default, every attempt in a hash-chained audit log. Single user, local only. v1.0.0.
go NOASSERTION ★ 1 -
Deny-by-default MCP Streamable HTTP projection for Hedron (Beta)
pypi MIT -
MCP server for the Pterodactyl panel: 105 tools, default-deny destructive gating, websocket console
MIT -
MCP server for CMS content operations with deny-by-default write scoping and audit output
MIT -
MCP server to operate a Proxmox cluster, with a deny-by-default classifier and out-of-band human approval
MIT -
bathygithub.com/russell0/bathyupdated this year
Agent-native network discovery. Typed MCP tools, evidence-backed findings, deny-by-default scope enforcement.
cargo Apache-2.0 -
Person roster MCP: who may DM you, see your space, joke with you — default deny. 人物名册与权限。
npm MIT -
Nix wrapper for Claude Code with hierarchical prompts, default-deny Bash, safety-default env injection, parameterized plugin loading via mkJuggler, and HTTP MCP server orchestration — read-only mirror of code.linenisgreat.com/clown
MIT -
MCP server that runs on an OpenWrt router: deny-by-default policy, audit log, rollback-armed config changes
MIT ★ 1 -
kennelboxgithub.com/datacarismo/kennelboxupdated this year
A kennel for your AI agents — firejail-sandboxed MCP workspace with default-deny command and file policies.
pypi MIT -
Hardened MCP server for Docmost — authenticated Streamable HTTP, 20 tools, default-deny writes. Works with Claude, Codex, Cursor.
MIT -
browser-agentgithub.com/tenmomo/browser-agentupdated this year
Local MCP browser broker and Chrome extension for coding agents — loopback-only, tab-scoped, finance-domain default-deny.
MIT -
Read-only Kubernetes access for AI agents over MCP, governed by deny-by-default policy.
MIT ★ 1 -
Agent-auditable security layer for npm — auditing proxy, deny-by-default install sandbox, provenance & signature verification, SBOM + signed attestations, MCP surface.
Apache-2.0 ★ 1 -
Drive your already-logged-in Chrome from any MCP client — Chrome extension + chrome.debugger, 74 tools, default-deny safety guard.
MIT -
petromcpgithub.com/ameyxd/petromcpupdated this year
Local-first MCP server for petroleum data. Read LAS well logs from Claude, behind a default-deny path allowlist.
pypi MIT ★ 3 -
Agent-native backend-as-a-service: a deny-by-default policy engine, self-correcting errors, and native MCP — in a single Go binary.
Apache-2.0 -
Local-first screen memory for AI agents — read your own screen history over stdio, with evidence anchors and a default-deny consent file.
Apache-2.0 ★ 1 -
Safety-gated QQ Zone MCP server: read / like / comment / publish with default-deny gates. Requires a OneBot gateway with QZone actions. QQ空间 MCP。
MIT ★ 2 -
rails-mcpgithub.com/jaimenbell/rails-mcpupdated this year
MCP server for default-deny action gating. classify_action always returns GATED; adds a spend-intent ledger and CLI sign-off trail. Registry ships empty by design.
pypi MIT -
Governed persistent AI workers with default-deny authority, human approvals, dry-runs, and an authenticated MCP.
Apache-2.0 -
MCP server exposing scoped filesystem operations (list, stat, read, glob, optional move/copy/delete) for media library organization. Path-whitelisted, deny-pattern-aware, dry-run-by-default writes.
MIT -
xrpl-hermesgithub.com/CarpXRPL/xrpl-hermesupdated this year
☤ Model-agnostic XRPL capability layer: certified L1 reads, unsigned builders, curated knowledge, and default-deny MCP. Keys stay yours.
MIT ★ 3 -
@envmanifest/mcp-localgithub.com/adam-scott-thomas/envmanifestupdated this year
Local MCP server exposing envmanifest contracts to coding agents (Claude Code, Cursor, etc). Names + metadata only, never values. Default-deny on mutating tools.
npm MIT -
lucygithub.com/ckaico/lucyactive
Give your AI the file, not the drive. One local index of Dropbox, iCloud Drive and attached drives, with a deny-by-default MCP gate.
Apache-2.0 -
Deterministic governance layer for AI agents: action-bound authority, deny-by-default with no LLM in the decision, and offline-verifiable Ed25519 receipts.
NOASSERTION ★ 9 -
Deny-by-default MCP server for GitLab: capability-based access for AI agents, a .noai content marker, and an authenticated git smart-HTTP proxy.
MIT ★ 1 -
Two hardened MCP servers - a terminal executor and a file organizer - on one shared deny-by-default safety core, with a fail-closed audit trail and undoable file moves.
MIT -
MCP server + Claude Code plugin for reviewing Bitbucket Cloud pull requests without cloning. Default-deny repo allowlist, separated write gates, no approve/merge tool by design.
MIT -
Security-hardened MCP server for terminal access: deny-by-default policy engine, shell=False sandboxed execution, cwd jail, secret redaction, fail-closed audit log.
MIT -
b2-mcp-servergithub.com/ffumero2003/b2-mcp-serverupdated this year
MCP server exposing Backblaze B2 Cloud Storage as tools an AI client can call — list, upload, download, delete, and report bucket usage against a budget. TypeScript, stdio, filesystem access fenced and denied by default.
★ 1 -
Use ChatGPT as the brain and your local computer as the hands — a safe, permissioned local MCP gateway with default-deny security, local approvals, and full audit logging.
npm MIT ★ 12 -
@evestack/mcpgithub.com/sammytourani/evestackupdated this year
evestack's control plane as an MCP server. Ask Claude Code why last night's agent run got denied, what it cost, and turn it into a regression test — over stdio, read-only by default.
npm Apache-2.0 -
Governed, identity-aware server for Agent Skills (SKILL.md): OIDC + default-deny policy per fetch, immutable digest-verified versions, audit log, native sync for every coding agent, and MCP (SEP-2640).
Apache-2.0 -
Multi-account Google Workspace (Gmail, Drive) for LLM agents — 100% local, default-deny policy, human-in-the-loop access. A macOS MCP server.
MIT -
Production Helm chart for SearXNG on Kubernetes — rootless, read-only rootfs, no init containers, deny-by-default NetworkPolicies, chart-managed credentials injected from Secrets, optional Valkey cache and MCP relay.
AGPL-3.0 ★ 4 -
Guardrails service for AI agents. Default-deny tool call evaluation with LLM safety analysis, priority-ordered decision matrix, and human-in-the-loop escalations. Session recording, behavioral analysis, MCP proxy, secret redaction, and real-time audit.
pypi NOASSERTION ★ 22 -
Read-only Kubernetes and AWS access for AI agents, behind a deny-by-default policy gate with a JSONL audit trail and eval suites that score both per-call block rate and whole-trajectory behaviour.
MIT -
Python library and MCP server that lets an AI agent act only within permission grants you write. Denies by default, re-checks each result, undoes failed writes, keeps a verifiable log.
pypi NOASSERTION ★ 3 -
MCP server for Home Assistant, transparency and security first: an untrusted-client gateway with default-deny policy, out-of-band approval, redaction and a hash-chained audit log. One static Go binary, runs as a Home Assistant App. Concept stage.
Apache-2.0 -
Drive your real Chrome browser over MCP — real logins, real cookies. A stdio MCP server (CLI) plus an MV3 extension, driving Chrome via chrome.scripting/chrome.tabs. Multi-tab batch automation, accessibility snapshots, deny-all security by default.
npm MIT -
pgguard-mcpgithub.com/Azzaraell/pgguard-mcpupdated this year
Read-only MCP server for Postgres. A deny-by-default policy file decides which tables and columns an AI agent can see; every query runs inside a BEGIN READ ONLY transaction with a statement timeout, and every decision lands in an audit log.
MIT -
CONTINUUM: Verifiable semantic recovery for long-running AI agents. Semantic checkpoints (not conversation dumps), an idempotent action ledger that refuses duplicate side effects, and a hash-chained tamper-evident event log, all exposed as a deny-by-default MCP server. Framework-agnostic, Python 3.11+
Apache-2.0 ★ 27 -
Local (stdio) MCP server in Go that gives opencode and other model hosts a safe, isolated sandbox for running arbitrary commands — a secure drop-in replacement for the host shell. Deny-by-default, ephemeral per-session containers with Docker, or macOS seatbelt sandbox-exec with no daemon. Streaming output.
go Apache-2.0 -
nostr-ops-mcpgithub.com/llmops-pro/nostr-ops-mcpno update data
Nostr operations for AI agents: query relays with NIP-01 filters, read profiles, verify NIP-05, NIP-19 encode/decode, publish notes / metadata / addressable events, and send/read encrypted DMs (NIP-44, NIP-04 fallback). Signs with a local nsec or a NIP-46 remote bunker so your key can stay off the machine. Safety: deny-by-default kind allowlist, DM allowlist, rate limiting, optional two-step confirm, audit log.
smithery