MCP servers, connectors and command line tools that let an agent reach other services and data.
46 entries. Latest source check .
What a source check covers
Documentation checked means the source was read. It does not mean the software was installed, tested or security audited.
Reading an entry does not authorise installation, credential access or spending. Tech Horizon Labs projects are labelled, and there are no affiliate links or paid placement.
People who want cited local search and agent memory without a hosted search cluster.
Access and use
Installer is curl -fsSL https://xerj.org/get | sh. Read the script before running it. The binary indexes local paths and can serve an Elasticsearch-compatible API on localhost. Apache-2.0.
Developers whose agents need to work on a Supabase project they control.
Access and use
Remote URL https://mcp.supabase.com/mcp with OAuth. Can list, query and change project data. Read Supabase's MCP security notes and restrict feature groups before connecting production.
People whose agents need live Notion context they already have permission to see.
Access and use
Remote URL https://mcp.notion.com/mcp with OAuth. On Business and Enterprise plans a workspace owner can also create an MCP token, which a client sends as a bearer token with no browser sign-in, and Enterprise workspaces with SAML SSO can use enterprise-managed authorization. Only workspace members can connect; guests cannot. Can read and write workspace content. Workspace owners can manage which clients connect. Prefer the hosted server over the unmaintained self-hosted repo.
People whose agents work on a Cloudflare account they control.
Access and use
Remote URL https://mcp.cloudflare.com/mcp with OAuth, or a scoped API token as a bearer. Can change account configuration. Grant the smallest permission set. Other product-specific Cloudflare MCP URLs exist; this is the main API server.
An MCP server and hooks layer that sandboxes large tool output, keeps a local session index, and can run analysis scripts so agents do not dump whole files into context.
People running Claude Code or other supported clients whose context windows fill with tool output.
Access and use
In Claude Code: /plugin marketplace add mksglu/context-mode then /plugin install context-mode@context-mode. MCP-only: claude mcp add context-mode -- npx -y context-mode. ctx_execute can run code on your machine. Elastic License 2.0: you may not offer it as a hosted service that exposes its features.
People whose coding agents spend context reading large files in Claude Code, Cursor, Codex or other MCP clients.
Access and use
Install with uv tool install jcodemunch-mcp, then jcodemunch-mcp init. Indexes live at ~/.code-index/. Free for personal use. Commercial use needs a paid licence. Default install sends an anonymous savings counter; set share_savings false to opt out.
A hosted MCP, CLI and skill for discovering, inspecting and running third-party data endpoints against one Monid balance. The public repo is the MIT Deno connector framework those endpoints are written in.
Teams who want one billed catalog of agent tools, or to add a connector as a pull request.
Access and use
Hosted MCP is https://mcp.monid.ai/v1 and signs in with a Monid account. Skill setup is the public file https://monid.ai/SKILL.md plus an API key from app.monid.ai. CLI is npm install -g @monid-ai/cli, then monid keys add. Hosted runs spend a prepaid balance or x402 USDC (a stablecoin payment), priced per call or per result on each endpoint. Local clone needs Deno 2.x; deno task check and deno task test run offline, and engine:run needs that provider's vendor key. MIT.
People whose coding agents need to read local documents as Markdown.
Access and use
npx -y @sylphx/anymd setup. Needs Node 18 or later. Also brew install sylphxai/tap/anymd, pip, Docker or cargo. Local MIT binary needs no API key. OCR and ASR model downloads need explicit opt-in. Confine the MCP server with --allow-dir. Hosted Sylphx AI parse is a paid metered service that uses the same crates. anymd Pro is an optional one-time purchase for cite-check and video timelines. MIT.
People who want many MCP servers behind one localhost endpoint, with a review gate before agents can call new tools.
Access and use
brew install --cask smart-mcp-proxy/mcpproxy/mcpproxy, or brew install smart-mcp-proxy/mcpproxy/mcpproxy for the CLI, or a GitHub Release for macOS, Windows or Linux. mcpproxy serve listens on 127.0.0.1:8080. Config and data live under ~/.mcpproxy/. New servers stay quarantined until you review and approve them. Update checks query GitHub Releases on startup and every 4 hours until you set update_check.enabled to false or MCPPROXY_DISABLE_AUTO_UPDATE=true. JavaScript code execution is on by default. The tray Unix socket is trusted without the API key until you set enable_socket to false. MIT.
People who want corrections and conventions to persist across Claude Code, Codex, Cursor, OpenCode and similar tools.
Access and use
npm install -g @plur-ai/cli@latest @plur-ai/mcp@latest, then plur init and plur doctor. Needs Node.js 20+. For Claude Code, init writes the MCP server into ~/.claude.json and hooks into ~/.claude/settings.json. plur doctor downloads a local search model when downloads are allowed; without it, search uses keywords. An optional openai-3-large embedder sends text to OpenAI. plur dashboard is loopback-only. plur sync to a personal git remote pushes private engrams; use a private repo. Apache-2.0.
People who want a coding agent to read Plausible stats with a Plausible API key.
Access and use
Hosted: claude mcp add --transport http plausible https://plausible-mcp.sentry.dev/mcp --header "Authorization: Bearer YOUR_PLAUSIBLE_API_KEY". Queries go to that Sentry-hosted server with your key. Tools are read-only. Local: Node.js 20+, git clone, pnpm install and pnpm build, then node dist/index.js with PLAUSIBLE_API_KEY. PLAUSIBLE_BASE_URL points at a self-hosted Plausible. The hosted /internal path is for Sentry staff. MIT.
A local MCP and REST server that turns English narration scenes into captioned short videos with Kokoro speech, Whisper captions, Pexels B-roll and Remotion.
People generating English short-form videos on their own machine or Docker host.
Access and use
Needs a free Pexels API key, about 4 GB RAM, 2 vCPU and 5 GB disk. Docker tiny image: docker run -it --rm --name short-video-maker -p 3123:3123 -e PEXELS_API_KEY= gyoridavid/short-video-maker:latest-tiny. Web UI and REST at http://localhost:3123. MCP SSE at /mcp/sse. Videos land in ~/.ai-agents-az-video-generator with npm, or /app/data in Docker. English voiceover only. Windows is unsupported. MIT.
Teams who want Claude, ChatGPT or Copilot to call existing APIs and ERPs through one MCP URL they host.
Access and use
Needs Docker 24+ and openssl. mkdir anythingmcp && cd anythingmcp, then curl the official docker-compose.quickstart.yml, write JWT_SECRET and ENCRYPTION_KEY with openssl rand -hex 32, then docker compose up -d. UI is http://localhost:3000; the first registered account is admin. MCP is http://localhost:4000/mcp. Keep the generated .env: ENCRYPTION_KEY decrypts stored connector credentials. Quickstart binds 127.0.0.1. Connector keys are stored encrypted and forwarded to the APIs you attach (SAP, shops, databases). SQL tools are read-only by default. The optional hosted trial at https://cloud.anythingmcp.com runs on the maintainers' servers in Frankfurt, Germany, and holds your connector credentials there. AGPL-3.0 for the open tree; ee/ directories use a separate commercial licence.
Testers who want an agent to drive a browser or a phone with do, get and check.
Access and use
MCP: curl -LsSf https://alumnium.ai/install.sh | sh (read the script before running it), then claude mcp add alumnium --env OPENAI_API_KEY=... -- alumnium mcp. Libraries: pip install alumnium, npm install alumnium, or the Java artifact. Agentic mode (default) sends page text, accessibility trees and screenshots to the model you set. Direct mode skips the model and lets your coding agent call driver actions. Browser profiles live in ~/.alumnium/profiles. MIT.
An open-source tool-calling platform that gives agents over 600 pre-built app integrations, with managed OAuth and secrets, natural-language permission boundaries and tool discovery, through a unified MCP server or Python and TypeScript SDKs.
Developers giving an agent or agentic IDE authenticated access to many SaaS tools, for themselves or for their own end users.
Access and use
Use the managed service at aci.dev with an account and API key, or run the platform locally: Python 3.12+, Docker Compose, PostgreSQL with pgvector and an OpenAI API key, with PropelAuth for the dev portal. Each linked app needs OAuth consent or API credentials from the account the agent will act on, held in ACI's secrets management. The unified MCP server is a separate repository, aipotheosis-labs/aci-mcp. Apache-2.0.
Xero's MCP server for reading and writing accounting and payroll data, including contacts, invoices, quotes, payments, bank transactions, manual journals, reports and timesheets.
Businesses and bookkeepers who want an agent to query or update a Xero organisation.
Access and use
Runs locally with Node.js 18 or later through npx -y @xeroapi/xero-mcp-server@latest. Authenticate with a Xero Custom Connection (XERO_CLIENT_ID and XERO_CLIENT_SECRET) or a bearer token from your own OAuth flow (XERO_CLIENT_BEARER_TOKEN). Xero's developer docs say Custom Connections are only for organisations in Australia, New Zealand, the UK and the US and need an additional monthly subscription, but connect to the Demo Company free for development. The tools can create invoices, payments, bank transactions and manual journals and approve timesheets, so narrow XERO_SCOPES and start on the Demo Company. Payroll queries need an NZ or UK organisation. MIT.
A Python MCP server, CLI and local dashboard that pulls markets, SEC filings, economic data, conflict, disaster, aviation, cyber and news feeds from public APIs, keeps a dated history, and can index results in Qdrant for semantic search.
Analysts and research agents that need a broad, cited view of world events and markets.
Access and use
Clone, pip install -e ., then run world-intel-mcp over stdio and add it to ~/.claude.json. The dashboard serves http://localhost:8501 and binds 127.0.0.1 by default. Most sources need no key; ACLED, UCDP, NASA FIRMS, EIA, FRED, Cloudflare, OpenSky, TomTom and Windy tools use optional credentials. The optional collector daemon fetches every source on a 5-minute default interval and stores history in ~/.cache/world-intel-mcp/cache.db. Qdrant runs separately in Docker, and AI briefs use a local Ollama. Each upstream source keeps its own terms of use, so check them before business use. MIT.
Marketers looking for creators whose audience matches their buyers.
Access and use
Needs an Influencers Club account with API credits, and each call spends them. A local install reads an API key from the environment, and its batch tools read and write local files. MIT licence.
Marketers and creators researching YouTube channels and content.
Access and use
Needs a vidIQ account, and hosted tools may use credits. It reads data for the channels you authorise, and some tools can make approved edits to videos you own. Apache 2.0 licence for the plugin.
Needs a Google Cloud project and an API key. The default quota is 10,000 units a day, and most reads cost 1 unit. Changes to a channel need its owner's OAuth consent.
Channel teams who want an agent to read and manage their own YouTube channel.
Access and use
Needs your own Google Cloud project and OAuth client, and it saves client_secret.json and youtube.token.json in the working folder unless you set its environment variables. The OAuth grant lets it create, update and delete videos, playlists and comments on that channel, and agent mode sends prompts to the Gemini or OpenAI key you supply. Data API quota applies; Apache 2.0 licence.
An MCP server with read-only tools for YouTube video, channel, playlist and transcript lookups, plus creator search filtered by subscriber count and recent uploads.
Marketers using an MCP client to shortlist creators by size and activity.
Access and use
Needs a YouTube Data API key in the environment, and each search spends that key's daily quota. It can fall back to a second and third key when one runs out, but the YouTube API Services Terms forbid getting round quota limits, so configure one key. Channel results can include emails found in public metadata, which are personal information; MIT licence.
An MCP server that fetches a YouTube video's captions, with optional timestamps, and by default strips sponsor and promotional sections marked by chapters.
People who want an agent to read or summarise what a video says.
Access and use
Runs through npx with no account or key for transcripts. It reads YouTube's caption data outside the official API, which YouTube's Terms of Service restrict, and the optional video analysis tool sends a direct video URL to TwelveLabs under your own TwelveLabs key. MIT licence.
HypeAuditor's hosted MCP server for YouTube, Instagram and TikTok creator analytics, including subscriber demographics, authenticity scores, growth and price estimates.
Brands and agencies vetting creators before paying for a campaign.
Access and use
Sign in with a HypeAuditor Pro or higher account; creator analytics spend AI credits at US$0.005 each, about 10 for a search and 100 for an analysed account, and the assistant asks before any paid unlock. It can also create media plans and move creators between campaigns in your account. HypeAuditor's terms bar bots and scraping outside its API, which is a separate product sold through its sales team.
A hosted transcript search service and MCP server for YouTube and podcasts that finds quotes, mentions, media appearances, ad reads and recurring channel sponsors.
Researchers tracking what is said about a brand and which brands sponsor which channels.
Access and use
Sign in through your MCP client or use an arc_sk_ account key; the free plan has 1,000 credits a month and withholds the newest 30 days, and sponsor and ad-read data needs the Pro+ plan. Reads cost credits per row, and the write tool can create and pause monitors but not delete them. Speaker and sponsor labels can be incomplete, and the plugin repository is under a proprietary licence that only permits running it.
Connects Claude or ChatGPT to 1of10's database of YouTube outlier videos, which beat their channel's usual views, and to its title and thumbnail tracking.
Creators and strategists looking for video ideas that already work in a niche.
Access and use
Needs a 1of10 Pro (US$89 a month) or Ultra plan, and you connect by signing in from its Claude and ChatGPT page; the server link for other clients appears only after sign-in. Research does not use credits but generating titles, ideas or thumbnails does, and it can save videos, create trackers and link your channel without deleting anything. There is no public API reference beyond the product page.
Check a domain's DNS-AID index record against DNS-AID draft-02, an IETF Internet-Draft that is still a work in progress. It checks DNS only, and a DNS-AID record does not change your Are you found by AI? score. It looks for an SVCB record (type 64) at _index._agents.yourdomain.com and does not test the endpoint the record points to, its TLS connection or whether an AI engine uses it.
People who operate an agent index and want to see whether their domain publishes its DNS-AID record.
Access and use
Free web form that asks only for a domain, with no sign-in. The page sends that domain to areyoufoundbyai.com, which looks up the record and shows the queried name, the target and whether the resolver reports an authenticated answer. The page says no DNS-AID setup is needed unless you operate an agent index, and to read its setup requirements at areyoufoundbyai.com/guides/dns-aid before changing DNS.
Microsoft's MCP server that lets an agent drive a real browser with Playwright, reading each page as an accessibility snapshot so it needs no screenshots or vision model.
Developers who want an agent to browse sites, fill in forms or test a web app from Claude Code, Codex, VS Code, Cursor or another MCP client.
Access and use
Runs locally with npx @playwright/mcp@latest, for example claude mcp add playwright npx @playwright/mcp@latest. Needs Node.js 18 or later. By default the browser keeps a persistent profile on disk (ms-playwright/mcp-* in your user cache folder), so sites you sign in to stay signed in; --isolated keeps the profile in memory instead. --extension connects to your running Chrome or Edge and its signed-in tabs. The README says the server is not a security boundary, and neither are --allowed-origins and --blocked-origins. --port serves it over HTTP on localhost; --host 0.0.0.0 opens it to the network. The README suggests coding agents may do better with the Playwright CLI and skills, which use fewer tokens. Apache-2.0.
Sentry's hosted MCP server that lets a coding agent look up Sentry issues, errors, traces and logs, and ask Sentry's Seer to analyse an issue, while you debug.
Developers whose apps report to Sentry and who want Cursor, Claude Code or another coding agent to pull in production errors while they fix them.
Access and use
Remote URL https://mcp.sentry.dev/mcp, for example claude mcp add --transport http sentry https://mcp.sentry.dev/mcp, which opens an OAuth sign-in to your Sentry account. Adding an organisation slug, or an organisation and project slug, to the URL path limits the session to that scope. Clients that send custom headers can pass a Sentry user token as Authorization: Sentry-Bearer instead. For self-hosted Sentry, run npx @sentry/mcp-server@latest over stdio with SENTRY_AUTH_TOKEN and SENTRY_HOST; the README's token scopes include project:write, team:write and event:write. When you run it yourself, the natural-language search tools need your own OpenAI, Azure OpenAI, Anthropic or OpenRouter key. The source moved from getsentry/sentry-mcp to getsentry/toolkit. The LICENSE.md file is FSL-1.1-Apache-2.0; GitHub shows NOASSERTION.
Atlassian's official hosted MCP server, which lets an agent search, summarise, create and update work in Jira, Confluence, Jira Service Management, Bitbucket, Compass and Loom.
Teams on Atlassian Cloud who want Claude, ChatGPT, Cursor, VS Code or a coding agent to work with their Jira issues and Confluence pages.
Access and use
Remote URL https://mcp.atlassian.com/v2/mcp (v2 is the recommended version), for example claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp, then /mcp to sign in. Needs an Atlassian Cloud site. Sign in with OAuth 2.1 or an API token; actions run with your existing Atlassian permissions. Jira Service Management tools work only with API tokens that an organisation admin has turned on, and the Jira delete and manage tools stay off until an admin enables them. Admins can choose which AI tools and domains may connect, and each tool call is recorded in the organisation's audit log. IP allowlists apply to MCP requests, so a hosted AI client calling from its provider's addresses may be refused. The repository's plugin files and skills are Apache-2.0; the server itself runs on Atlassian Cloud.
Developers and Stripe account owners who want an agent to build a Stripe integration, look up Stripe data or analyse subscriptions and billing.
Access and use
Remote URL https://mcp.stripe.com. npm install -g @stripe/cli@latest then stripe agent setup configures it, with Stripe's skills, for the agents it detects. Sign in with OAuth, choosing which live accounts or sandboxes the client may use, or pass an agent API key as a bearer token. From December 2026 it stops accepting full-access secret keys and restricted keys without the Agent tag. The stripe_api_write tool can call any Stripe API POST, PATCH, PUT or DELETE method. Some actions, such as refunds, wait for you to approve them in the Dashboard, and Stripe asks you to turn on tool approval in your client as well. Admins can turn MCP access on or off for live mode and for sandboxes. The older @stripe/agent-toolkit and @stripe/mcp packages are deprecated and get no security fixes. Stripe's plugin and skills repository, github.com/stripe/ai, is MIT.
Teams on Linear who want Claude, Cursor, Codex or another agent to read and update their issues and projects.
Access and use
Remote URL https://mcp.linear.app/mcp over Streamable HTTP, for example claude mcp add --transport http linear-server https://mcp.linear.app/mcp, then /mcp to sign in with OAuth 2.1. That endpoint can read and write. For read-only access, connect to https://mcp.linear.app/mcp/readonly or request only the read OAuth scope. A Linear API key or OAuth token can go in an Authorization: Bearer header instead; a key with only the Read permission keeps it read-only. Each workspace needs its own sign-in. Clients without remote MCP support can use npx -y mcp-remote https://mcp.linear.app/mcp. The /sse endpoint is deprecated.
Figma's MCP server, which gives coding agents the design context in Figma files, such as variables, components and layout, for generating code, and lets agents write native content back to the canvas.
Designers and developers who turn Figma designs into code with Claude Code, Codex, Cursor, VS Code or another MCP client.
Access and use
Remote server at https://mcp.figma.com/mcp, for example claude mcp add --transport http figma https://mcp.figma.com/mcp, or Figma's plugin with claude plugin install figma@claude-plugins-official. You sign in with your Figma account and can reach only files you already have access to. Starter plans and View or Collab seats get up to 20 read tool calls a month; Full and Dev seats on paid plans get 200 to 600 a day. Writing outside your drafts needs a Full seat. Writing to the canvas is free during the beta and will become a usage-based paid feature. generate_image and send_prompt use Figma AI credits. A desktop server that runs through the Figma desktop app needs a Dev or Full seat. Use falls under the Figma Developer Terms.
Slack's hosted MCP server, with tools to search a workspace, read channels and threads, send and schedule messages, and manage canvases, lists and files.
Slack workspaces that want Claude, Cursor or Perplexity to use their Slack history, and developers building an MCP client on a Slack app.
Access and use
Ready to use in Claude.ai, Claude Code, Perplexity and Cursor. A client you build must be backed by a registered Slack app, either internal or published in the Slack Marketplace, and connects to https://mcp.slack.com/mcp over Streamable HTTP with confidential OAuth (the app's client ID and secret, or PKCE for desktop clients). SSE and dynamic client registration are not supported. Workspace admins approve and manage MCP client apps. User token scopes follow the tools, for example search:read.public to search, chat:write to send and channels:history to read. MCP activity appears in Slack's audit logs, the app's IP allowlist applies, and standard Slack API rate limits apply per tool.
Zapier's hosted MCP server that lets an AI client run actions in the 9,000 or more apps Zapier connects to, such as sending a Slack message or creating a task, through your Zapier account.
Zapier users who want Claude, ChatGPT, Cursor, VS Code or another MCP client to act in their other apps without an integration for each one.
Access and use
Clients connect to https://mcp.zapier.com/api/v1/connect, and most sign in with OAuth. Plugins for Claude Code, Codex, Cursor and Gemini CLI come from github.com/zapier/zapier-mcp (MIT). Actions run through the same app connections as your Zaps, so the agent acts with those accounts. In agentic mode the AI adds the actions it needs during a conversation; managed mode limits it to a fixed set you pick. Each successful tool call uses two tasks from your Zapier plan. Sharing a server needs a Team or Enterprise plan. Clients must support Streamable HTTP; SSE is not supported.
Developers whose coding agents write Gemini API code and need current model names, request schemas and limits.
Access and use
Remote URL https://gemini-api-docs-mcp.dev. Claude Code: claude mcp add --transport http geminiDocs https://gemini-api-docs-mcp.dev, or npx add-mcp https://gemini-api-docs-mcp.dev for the agents it detects. Two tools, gemini_search_docs and gemini_get_doc, both over Google's public Gemini API documentation. Google's setup page uses no key or sign-in for it and names no price. The same documentation is also listed for agents at https://aistudio.google.com/llms.txt.
Google's MCP server for searching and reading its official developer documentation, including Firebase, Google Cloud, Android and Google Maps Platform.
Developers whose coding agents build on Google products and need current documentation.
Access and use
Remote URL https://developerknowledge.googleapis.com/mcp. Enable the Developer Knowledge API in a Google Cloud project, then send an API key in the X-Goog-Api-Key header, or use OAuth or Application Default Credentials. Tools: search_documents, get_documents and answer_query, which writes an answer from the documentation. Default quotas per project are 100 searches and 100 document reads a minute, and 50 generated answers a day. The setup and quota pages name no price.
Google Maps Platform's MCP server, formerly Maps Grounding Lite, that lets an agent search places, look up weather, compute routes and turn place names or Maps links into place IDs.
Teams building agents that answer questions about places, routes, travel times or local weather.
Access and use
Remote URL https://mapstools.googleapis.com/mcp with a Google Maps Platform API key in the X-Goog-Api-Key header, or OAuth. Five tools: search_places, lookup_weather, compute_routes, resolve_names and resolve_maps_urls. Essentials requests are free for the first 10,000 a month, then US$7.00 per 1,000. Pro options cost US$14.00 per 1,000 after 5,000 free a month, and for now work only on allowlisted projects. Search, weather and routes each allow 300 queries a minute per project. Answers must show the Google Maps sources each response returns, and the model you use must not cache, store or train on Google Maps content. The Maps Demo Key covers it for prototyping.
Microsoft's MCP server, in preview, that turns plain questions about a Microsoft Entra tenant's users, groups, apps and devices into read-only Microsoft Graph calls.
IT admins and helpdesk staff who want an agent to answer identity and directory questions about their own tenant.
Access and use
Remote URL https://mcp.svc.cloud.microsoft/enterprise, in preview under the Microsoft APIs Terms of Use. An Application Administrator or Cloud Application Administrator registers it once per tenant with the Microsoft.Entra.Beta PowerShell module and grants MCP scopes to each client; Claude and other non-Microsoft clients need their own app registration. Its Graph tool, microsoft_graph_get, is read-only and honours the signed-in user's roles and the scopes granted. The server has no extra cost or licence, but you need licences for the data you read. Limit: 100 calls a minute per user, plus Graph throttling. Global cloud only. Turn on Microsoft Graph activity logs to record every call.
OpenRouter's hosted MCP server that lets a coding agent look up live models, prices, benchmarks, providers and your credit balance, and send test messages to any model.
Developers choosing or testing models on OpenRouter from inside their coding agent.
Access and use
Remote URL https://mcp.openrouter.ai/mcp. Claude Code: claude mcp add --transport http openrouter https://mcp.openrouter.ai/mcp, then claude mcp login openrouter. Signing in with OAuth creates a separate key for this connection that expires within 7 days and starts with a US$10 spend cap, which you can change on the approval screen. Most of its 17 tools are read-only lookups; send-message and generate-image spend credits, and send-feedback sends a report on one of your generations to OpenRouter. OpenRouter suggests it for building and testing, and the API for running models in your app.