MCP servers · shortlist
MCP servers for developers
Source control, continuous integration, code execution and package tooling — the servers that put an agent inside a real build pipeline rather than beside it.
How this list was ordered. There is no reliable popularity signal in the source registries, so entries are ranked by how completely the record is documented — a substantial description plus a documentation, repository or base URL — and then alphabetically. 531 mcp servers in this scope clear the content bar; the top 24 are shown. A place here is not an endorsement or a paid slot.
Shortlisted entries
- 01
JeanFrancoisGagne/crapkit
Read-only view of a repository's CRAP scores, one per function: ccn^2 (1 - cov)^3 + ccn, cyclomatic complexity joined to the branch coverage your own test command already writes. Twelve tools (getnextitem, listworklist, getfunctionbrief, getfunctionhistory, checkgate, checkconfig, listcoupledfiles, listduplicatefunctions, getratchetreport, gettrend, listruns, listclaims) hand the agent the worst function to fix next with its own source, its uncovered lines, its change-coupled files, its near-duplicates and a gate verdict on the edited file. None of the twelve runs a test suite or edits source files. Reads Python, TypeScript, JavaScript, Go, Rust, Swift, Java, C/C++ and…
Developer Tools auth: None
- 02
ARADIA-systems/aradia-mcp-server
Turnkey on-premise private Agentic AI systems by ARADIA on dedicated NVIDIA DGX hardware (Spark, Station, B200) with zero cloud data leakage. Operating since 1991, six tools cover hardware sizing (queryhardwarespecs), CapEx token ROI payback modeling (calculateroi), autonomous affiliate onboarding to earn 10% bounties (onboardpartner), and machine procurement via BTCPay Server (BTC/LN/USDT) or Stripe MPP Fiat (executeprocurementorder), plus trackorderstatus and querypartnerstats. Listed on the official MCP Registry (com.aradia/sovereign-hardware). Remote Streamable HTTP / SSE at https://aradia.com/api/mcp or https://aradia.com/api/a2a/mcp-endpoint.php.
Developer Tools auth: API Key
- 03
cq27-dev/rag-rat
Local repo-intelligence index and MCP server for coding agents — "what a repository knows about itself." Hybrid BM25 + vector semantic search, a tree-sitter code graph (callers/callees/imports across Rust, TS/TSX, Kotlin, C/C++, Python) with an optional compiler-grade SCIP oracle, queryable git/GitHub history, and typed, source-anchored repo memories (Invariant, Decision, Risk, …) that persist across sessions. Every result carries provenance — confidence labels and coverage warnings. Keeps source read-only; a background watcher keeps queries fresh. PreToolUse hooks augment grep and flag near-duplicate code as you write it. cargo install rag-rat
Developer Tools auth: None
- 04
carloshpdoc/memorydetective
iOS leak hunting and performance investigation. Reads .memgraph (leaks(1)) and .trace (xctrace) files, classifies retain cycles against a 34-pattern catalog (SwiftUI, Combine, Swift Concurrency, UIKit, Core Animation, Core Data, SwiftData, RxSwift, Realm), surfaces a Swift fixTemplate snippet for each match, and bridges to source via SourceKit-LSP. Single-call build+boot+launch fixes the macOS 26.x leaks --outputGraph regression; replayScenario + captureScenarioState close the verify-fix loop. 31 MCP tools, 34 catalog resources, 5 investigation prompts. CI-gateable via verifyFix and compareTracesByPattern. Apache 2.0.
Developer Tools auth: None
- 05
WCAG-Compliance/wcagc-mcp
Deterministic axe-core accessibility scanning (WCAG 2.1 AA, EN 301 549, PDF/UA) through a wcagc account: scan any public URL or PDF, crawl a whole registered site, replay saved multi-step journeys, and read violation trends. Returns structured findings — rule, severity counts and the success criteria involved — so the assistant reasons over real results rather than guessing from markup. Every scan result carries an automated-coverage disclaimer instead of a score or a pass/fail verdict. Hosted at https://mcp.wcagc.com/mcp, or run it locally with npx -y @wcagc/mcp.
Developer Tools auth: None
- 06
dwgx/SmartCLI
Drive full-screen terminal programs (vim, htop, lazygit) by reading the screen, not the byte stream: a live cell grid exposes which row is reverse-video selected, cursor position and alt-screen state, so waits are screen-state predicates instead of sleeps. The emulation is differentially tested against real tmux and GNU screen (cell-by-cell grid diffs plus a generative VT fuzz), which caught 12 emulation bugs including missing alternate-screen support. pip install smartcli-toolkit then smartcli-mcp, or uvx --from smartcli-toolkit smartcli-mcp.
Command Line auth: None
- 07
kao273183/mk-qa-master
AI 測試大師 — end-to-end QA loop over MCP. Drives pytest / Jest / Cypress / Go / Maestro from one surface; analyzes URLs (Web DOM) or live mobile screens (maestro hierarchy) to extract testable modules, generates runnable pytest or Maestro YAML with real selectors (not # TODO stubs), runs with auto-retry, then writes a prioritized optimization-plan.md ranked by evidence (flaky vs broken vs coverage gap). Mobile-first: iOS Simulator, Android Emulator, real devices, and BlueStacks (QAANDROIDHOST=127.0.0.1:5555). Install via uvx mk-qa-master.
Developer Tools auth: None
- 08
20000419/fauxnix
bash for AI agents on Windows via deterministic bash→PowerShell translation — no WSL, VM, or bash toolchain. MCP bash tool with GNU-style output, bash-style errors, coreutils exit codes, session-persisted cwd/env across calls, and per-file UTF-8/GBK sniffing (GBK files grep correctly, unlike Git Bash). Unsupported bash constructs fail loudly with the construct named; native commands (git/node/npm) pass through with argv quoting. npm install -g fauxnix-cli; verified with Claude Code, Codex, OpenCode, Kimi Code and Qwen Code.
Command Line auth: None
- 09
forgemeshlabs/aso-audit-mcp
Free Agent Signal Optimization scanner for website/API agent readiness. Scores discovery, identity, trust, commerce, reputation, and memory signals; validates robots.txt AI bot rules, llms.txt, agent.json, Google A2A agent cards, MCP server cards, API catalogs, OAuth metadata, x402, pricing, security.txt, status endpoints, and cross-file identity consistency. Returns ASO maturity levels, evidence, and prioritized fix plans. Local stdio or Glama-hosted, no API key required. npx -y @forgemeshlabs/aso-audit-mcp.
Developer Tools auth: API Key
- 10
sebastienrousseau/rlg
Log-stream tools for on-call / SRE agent workflows — expose rlg (RustLogs) log files as MCP tools. Three tools: taillog (last N events, filtered by level/target), filterlog (JSON-path-style predicates), summarizeerrors (grouped error taxonomy). Speaks JSON-RPC 2.0 over stdio; near-lock-free structured logging engine underneath (65k-slot ring buffer, 14 output formats, oslog/journald sinks). Install cargo install rlg-mcp or docker run -i ghcr.io/sebastienrousseau/rlg-mcp. Dual-licensed MIT OR Apache-2.0.
Developer Tools auth: None
- 11
MikhailHal/ariadne
Affected-test selection for Kotlin/Android projects: returns only the unit tests impacted by the current working-tree changes, so agents skip full test runs while iterating. Function-level static analysis via Kotlin Analysis API — call graph over functions, properties (initializers, delegates, accessors), constructors and callable references — with automatic multi-module and Android variant source-set discovery. Verified on Now in Android (34 modules, ~4s per query). brew install mikhailhal/tap/ariadne
Developer Tools auth: None
- 12
srbsa/diffgate
Deterministic guardrail for AI-written diffs, runnable from inside the coding agent's own loop over MCP. Zero-LLM-token, AST-precise rules (SQLi, SSRF, XXE, permissive CORS, prototype pollution, hardcoded secrets) across 8 languages, plus structural checks for reinvented helpers and over-abstraction — findings scoped to only the changed lines, same input always gives the same output, no false-block guarantee. npm i -g diffgate-review (registers as the diffgate command; diffgate mcp starts the server)
Developer Tools auth: API Key
- 13
saiffmirza/kiyas
Design-fidelity checks for AI agents: compare a Figma frame (or any design screenshot) against the rendered implementation and get structured discrepancies with CSS properties and severity. Resolves a natural-language component description to a URL + selector in your codebase, screenshots it with Playwright, and compares via vision AI using your existing Claude Code or Codex subscription — no API keys. 90% mutation recall, zero false positives on its golden eval set. npx -y kiyas-cli mcp
Developer Tools auth: API Key
- 14
gerard-kanters/mcp-linux-tools
Administer a Linux server from an MCP client such as Cursor or VS Code: read files and tail logs, check/reload/restart systemd services, manage cron jobs inside a dedicated crontab section, run read-only MySQL queries, drive WP-CLI on WordPress sites, run safe Git commands, and execute sandboxed Python. Directories, services and WordPress sites are allowlisted in config.json, and every tool returns a uniform {success, data, error, meta} response. Runs as a systemd service over HTTP.
Command Line auth: None
- 15
pofky/asc-mcp
App Store Connect control plane: 41 job-shaped tools that edit version metadata (with Apple's character limits validated), upload screenshots, build/sign/upload the binary, drive TestFlight, create IAPs and subscriptions with territory availability, run a release preflight audit, submit and release. Flags the steps Apple only allows on its website (privacy label, EU trader, first IAPs) instead of failing on them. Six tools need no license, three of those no Apple credentials either.
Developer Tools auth: None
- 16
kyisaiah47/kynth-mcp
Eleven keyless read-only lookups for questions a model cannot answer from a training cutoff, because the fact changed after it: what a given AI model costs today, whether a dependency is still maintained, what a SaaS stack costs per month, item-level search across public shadcn registries, published agent skills and configs, AI app-builder App Store readiness, ADA Title II accessibility reports for US .gov sites, and nonprofit good-standing. No API key, no signup. npx -y kynth-mcp
Developer Tools auth: API Key
- 17
shakaran/symfony-agent-mcp
Read-only introspection of a Symfony codebase: routes, controllers, services, entities, database schema, migrations, events, forms, security, Doctrine, Messenger, Twig and API Platform. 1,677 tools discovered progressively — tools/list starts at five meta-tools and the rest activate per category, so connecting costs ~828 tokens instead of ~154,000. Nothing is executed and nothing is written; results are scanned for credentials before they leave. npx -y @shakaran/symfony-agent-mcp
Developer Tools auth: API Key
- 18
trakshan-mishra/Diffcontext
Static-analysis-powered context compiler for LLM coding agents. Parses Python repos into dependency graphs, selects the minimal code context for a change (callers, callees, related functions), and packs it into a token budget with a disclosure header. Optionally biases retrieval from the bug report text. Includes verify which mines your repo's git history and grades retrieval — printing NULL RESULT when it doesn't fit. Zero runtime dependencies. pip install "diffcontext[mcp]".
Developer Tools auth: API Key
- 19
Komdosh/komnet
Git-backed message bus letting coding agents on different machines coordinate: rooms are folders, messages are files, git history is the log, and there is no server — the transport is a private git repository you already own. Questions, answers, decisions, collaborative tasks with claiming and recovery, and delegated repository reviews prepared in isolated worktrees at pinned revisions. Items marked needs: human cannot be answered by an agent, only relayed to a person.
Coding Agents auth: None
- 20
dbhq-uk/heliograph
Run commands on a machine you cannot SSH into. Publishes a step to a transport the far side already reaches (git, file share, S3-compatible storage, relay or a carried file); the whole run comes back as a log with every line timestamped in UTC, whether it passed or failed. For air-gapped, bastion-only and change-controlled estates: no tunnel, no proxy, nothing held open. The gates live on the far side, so a tool call can ask but cannot authorise. heliograph mcp.
Command Line auth: None
- 21
cygnussystems/cygnus-ssh-mcp
MCP server for SSH-based control of remote Linux, macOS, and Windows servers with 46 purpose-built tools instead of one generic command-runner: line-level file editing (no download/edit/upload round-trip), background task launch/monitor/kill, real sudo support, host aliases, recursive directory ops, archive create/extract, and full Unicode via SFTP (avoids PowerShell OEM code page corruption on Windows targets). pip install cygnus-ssh-mcp or uvx cygnus-ssh-mcp.
Command Line auth: None
- 22
andreilungeanu/cursor-delegate-mcp
Delegate multi-file implementation from Claude Code, Codex, or Copilot CLI to Cursor's CLI agent (Composer 2.5) over ACP. Returns a typed result — final message, agent-reported changed files, session id, and plan — instead of scraped terminal output. Plan mode drafts first, then resumes the same session to implement; clarifying questions come back as normal results rather than stalling. Ships plugin manifests for Claude Code, Codex, and Copilot. MIT.
Coding Agents auth: None
- 23
2ools/mcp-server
Turn what your assistant just built into a durable, versioned project. One free authless tool (createpublicpreview) renders a standalone HTML document as a real shareable URL; 42 more cover projects, refinement, version comparison, separately-delegated approvals, artifacts, work assignments and source export. Build authority and review authority are deliberately separate scopes. Remote Streamable HTTP, OAuth 2.1 + PKCE. https://2ools.app/mcp
Developer Tools auth: OAuth
- 24
no-materials/hird
Compiler-as-MCP-server for the Hirð language (typed actors and effect rows on the BEAM): type inference, definition lookup, effect-row explanation, IR fragments, actor protocol and supervision graphs, and token-budget-aware symbol summaries, so agents query the compiler instead of reading source. Errors come back structured, so an agent can self-correct from tool output alone. cargo install --git https://github.com/no-materials/hird hird-mcp
Developer Tools auth: API Key
Snapshot 2026-09-18. Fields shown are what the source registry stated, not a live probe — check the linked documentation before integrating. See the methodology.