---
title: "MCP platforms"
description: "Compare MCP platforms: the official MCP Registry, directories like Glama and Smithery, hosted tools like Composio, and gateways and tools for MCP servers."
type: "platform-category"
canonical_url: "https://theaiengineering.com/platforms/mcp-platforms"
markdown_url: "https://theaiengineering.com/platforms/mcp-platforms.md"
topic: "Model Context Protocol"
author: "The AI Engineering Editors"
publisher: "The AI Engineering"
citation: "The AI Engineering Editors. \"MCP platforms.\" The AI Engineering. https://theaiengineering.com/platforms/mcp-platforms"
---

# MCP platforms

MCP platforms are the registries, hosting services, integration platforms, gateways, and developer tools that help teams find, run, secure, and test Model Context Protocol servers, the programs that expose tools and data to AI agents and assistants.

The official MCP Registry, still in preview, publishes metadata for public servers for downstream aggregators, while directories such as the GitHub MCP Registry, Glama, PulseMCP, and MCP.so add search and curation. Docker packages servers as container images, Cloudflare and Vercel host remote servers, and Composio, Pipedream, Zapier, and Arcade supply authenticated tools for thousands of apps.

Gateways such as Bifrost, ToolHive, ContextForge, Metorial, and Speakeasy's MCP gateway put access control and audit logs in front of many servers, and FastMCP, mcp-use, MCP Inspector, and MCPJam help developers build and debug servers. Smithery, now part of Arcade, shut down its hosting in October 2026 and keeps its public registry.

## MCP platforms compared

| Platform | Open source | Deployment | Best for |
| --- | --- | --- | --- |
| [Arcade](https://www.arcade.dev) | No | Managed cloud | Governed, user-authorized tool calls for enterprises |
| [Bifrost MCP Gateway](https://docs.getbifrost.ai/mcp/overview) | Yes (Apache-2.0) | Self-hosted | One gateway for model calls and MCP tool calls |
| [Cloudflare remote MCP servers](https://developers.cloudflare.com/agents/model-context-protocol/) | No | Managed cloud | Public remote MCP servers with OAuth |
| [Composio](https://composio.dev) | No | Managed cloud, API | Multi-tenant agents acting in users' app accounts |
| [ContextForge](https://ibm.github.io/mcp-context-forge/) | Yes (Apache-2.0) | Self-hosted | Self-hosted gateway for MCP, A2A, and REST tools |
| [Docker MCP Catalog and Toolkit](https://docs.docker.com/ai/mcp-catalog-and-toolkit/) | No | Desktop app, Local | Running MCP servers as isolated containers |
| [FastMCP](https://gofastmcp.com) | Yes (Apache-2.0) | Library | Building MCP servers in Python |
| [GitHub MCP Registry](https://github.com/mcp) | No | Managed cloud | Developers who discover tools on GitHub |
| [Glama](https://glama.ai/mcp/servers) | No | Managed cloud | Finding and vetting open-source MCP servers |
| [Manufact](https://manufact.com) | Yes (MIT) | Library, Managed cloud | Shipping MCP apps to ChatGPT and Claude |
| [MCP Inspector](https://modelcontextprotocol.io/docs/tools/inspector) | Yes (Apache-2.0 AND MIT) | Local, CLI | Debugging an MCP server during development |
| [MCP Market](https://mcpmarket.com) | No | Managed cloud | Finding MCP servers and agent skills together |
| [MCP on Vercel](https://vercel.com/docs/mcp) | No | Managed cloud | Adding an MCP endpoint to an existing web app |
| [MCP Registry](https://registry.modelcontextprotocol.io) | Yes (Apache-2.0 AND MIT) | API, Self-hosted | Publishing a server once for every MCP directory |
| [MCP.so](https://mcp.so) | No | Managed cloud | Browsing community MCP servers by category |
| [MCPJam](https://www.mcpjam.com) | Yes (Apache-2.0) | Local | Testing MCP servers and OAuth flows in CI |
| [Metorial](https://metorial.com) | No | Managed cloud, Self-hosted | Company-wide MCP access behind single sign-on |
| [Pipedream](https://pipedream.com) | No | Managed cloud | Embedding user-authorized app tools in agent products |
| [PulseMCP](https://www.pulsemcp.com) | No | Managed cloud | Following MCP ecosystem changes |
| [Smithery](https://smithery.ai) | No | Managed cloud | Looking up vendor MCP endpoints |
| [Speakeasy MCP Gateway](https://www.speakeasy.com/product/mcp-gateway) | Yes (AGPL-3.0) | Managed cloud, Self-hosted | Governed MCP access for employees and agents |
| [ToolHive](https://stacklok.com/platform/) | Yes (Apache-2.0) | Self-hosted, Local | Platform teams running MCP servers on Kubernetes |
| [Zapier MCP](https://zapier.com/mcp) | No | Managed cloud | Giving assistants actions in many SaaS apps |

## Profiles

### Arcade

Arcade is an actions runtime for enterprise AI agents that enforces authorization, executes tool calls, and governs agent actions from one control point, with an MIT-licensed framework for building MCP servers; Arcade acquired Smithery in 2026.

- Company: Arcade.dev
- Website: https://www.arcade.dev
- GitHub: https://github.com/ArcadeAI/arcade-mcp
- Open source: No
- Deployment: Managed cloud
- Best for: Governed, user-authorized tool calls for enterprises
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#arcade-dev

### Bifrost MCP Gateway

Bifrost MCP Gateway is the MCP layer of Maxim AI's open-source Bifrost AI gateway, exposing connected MCP servers to hosts such as Claude Desktop and Cursor at one /mcp endpoint, with per-virtual-key tool filtering, OAuth, and a Code Mode that cuts tool-definition tokens.

- Company: Maxim AI
- Website: https://docs.getbifrost.ai/mcp/overview
- GitHub: https://github.com/maximhq/bifrost
- Open source: Yes (Apache-2.0)
- Deployment: Self-hosted
- Languages: Go
- Best for: One gateway for model calls and MCP tool calls
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#bifrost-mcp-gateway

### Cloudflare remote MCP servers

Cloudflare lets developers build and deploy remote MCP servers on Workers with its Agents SDK, using Streamable HTTP transport and OAuth so MCP clients can be authorized to act on a user's account.

- Company: Cloudflare
- Website: https://developers.cloudflare.com/agents/model-context-protocol/
- Open source: No
- Deployment: Managed cloud
- Languages: TypeScript
- Best for: Public remote MCP servers with OAuth
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#cloudflare-remote-mcp

### Composio

Composio is an integration platform for agents that handles authentication, tool discovery, and execution across a large catalog of app toolkits, available through Python and TypeScript SDKs and MCP, for products where each user connects their own accounts.

- Website: https://composio.dev
- GitHub: https://github.com/ComposioHQ/composio
- Open source: No
- Deployment: Managed cloud, API
- Languages: Python, TypeScript
- Best for: Multi-tenant agents acting in users' app accounts
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#composio

### ContextForge

ContextForge is IBM's open-source AI gateway that federates MCP servers, A2A agents, and REST or gRPC APIs behind one control plane with centralized governance, discovery, and observability.

- Company: IBM
- Website: https://ibm.github.io/mcp-context-forge/
- GitHub: https://github.com/IBM/mcp-context-forge
- Open source: Yes (Apache-2.0)
- Deployment: Self-hosted
- Languages: Python
- Best for: Self-hosted gateway for MCP, A2A, and REST tools
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#contextforge

### Docker MCP Catalog and Toolkit

Docker's MCP Catalog on Docker Hub lists verified MCP servers packaged as container images with versioning, provenance, and security updates, and the MCP Toolkit lets developers configure servers once and connect several AI clients to them.

- Company: Docker
- Website: https://docs.docker.com/ai/mcp-catalog-and-toolkit/
- Open source: No
- Deployment: Desktop app, Local
- Best for: Running MCP servers as isolated containers
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#docker-mcp-catalog

### FastMCP

FastMCP is an open-source Python framework for building MCP servers, clients, and interactive applications, maintained by Prefect, which also sells FastMCP Gateway for hosting servers with a private registry, tool-level access control, and audit trails.

- Company: Prefect
- Website: https://gofastmcp.com
- GitHub: https://github.com/PrefectHQ/fastmcp
- Open source: Yes (Apache-2.0)
- Deployment: Library
- Languages: Python
- Best for: Building MCP servers in Python
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#fastmcp

### GitHub MCP Registry

The GitHub MCP Registry is GitHub's searchable catalog of MCP servers from vendors and the community, with an install option on each listing for connecting servers to agents and editors that use MCP.

- Company: GitHub
- Website: https://github.com/mcp
- Open source: No
- Deployment: Managed cloud
- Best for: Developers who discover tools on GitHub
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#github-mcp-registry

### Glama

Glama runs a searchable registry of MCP servers, connectors, clients, and tools ranked by security, compatibility, and ease of use, alongside MCP hosting, an MCP gateway, a browser-based chat client, an inspector, and a directory API.

- Website: https://glama.ai/mcp/servers
- Open source: No
- Deployment: Managed cloud
- Best for: Finding and vetting open-source MCP servers
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#glama-mcp

### Manufact

Manufact, formerly mcp-use, is a cloud platform for deploying, testing, observing, and publishing MCP servers and MCP apps, built by the team behind the MIT-licensed mcp-use SDK for building MCP servers and clients.

The company and cloud product were renamed from mcp-use to Manufact, and the open-source SDK keeps the mcp-use name.

- Website: https://manufact.com
- GitHub: https://github.com/mcp-use/mcp-use
- Open source: Yes (MIT)
- Deployment: Library, Managed cloud
- Languages: TypeScript, Python
- Best for: Shipping MCP apps to ChatGPT and Claude
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#manufact

### MCP Inspector

MCP Inspector is the Model Context Protocol project's open-source developer tool for testing and debugging MCP servers interactively in the browser, on the command line, and in the terminal.

- Website: https://modelcontextprotocol.io/docs/tools/inspector
- GitHub: https://github.com/modelcontextprotocol/inspector
- Open source: Yes (Apache-2.0 AND MIT)
- Deployment: Local, CLI
- Languages: TypeScript
- Best for: Debugging an MCP server during development
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#mcp-inspector

### MCP Market

MCP Market is a directory of MCP servers and agent skills that connect clients such as Claude and Cursor to tools like Figma, Databricks, Storybook, and Ghidra, organized for browsing by category.

- Website: https://mcpmarket.com
- Open source: No
- Deployment: Managed cloud
- Best for: Finding MCP servers and agent skills together
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#mcp-market

### MCP on Vercel

Vercel hosts MCP servers as functions using mcp-handler, a package built on the MCP TypeScript SDK that turns a server definition into a Web-standard request handler for Next.js, Nuxt, SvelteKit, Hono, and other frameworks.

- Company: Vercel
- Website: https://vercel.com/docs/mcp
- GitHub: https://github.com/vercel-labs/mcp-handler
- Open source: No
- Deployment: Managed cloud
- Languages: TypeScript
- Best for: Adding an MCP endpoint to an existing web app
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#vercel-mcp

### MCP Registry

The MCP Registry is the Model Context Protocol project's official metadata repository for publicly accessible MCP servers, with an open API that downstream marketplaces and private registries consume; it stores server metadata rather than code and is in preview.

- Website: https://registry.modelcontextprotocol.io
- GitHub: https://github.com/modelcontextprotocol/registry
- Open source: Yes (Apache-2.0 AND MIT)
- Deployment: API, Self-hosted
- Languages: Go
- Best for: Publishing a server once for every MCP directory
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#official-mcp-registry

### MCP.so

MCP.so is a community directory of MCP servers for Claude, Cursor, and other AI clients, with server listings, remote endpoints, and setup guides grouped by category such as coding, browsers, and databases.

- Website: https://mcp.so
- Open source: No
- Deployment: Managed cloud
- Best for: Browsing community MCP servers by category
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#mcp-so

### MCPJam

MCPJam is an open-source inspector that runs locally for testing, debugging, and evaluating MCP servers, with OAuth flow testing, LLM-powered evals, and CI/CD integration; a few evaluation components use a separate license.

- Website: https://www.mcpjam.com
- GitHub: https://github.com/MCPJam/inspector
- Open source: Yes (Apache-2.0)
- Deployment: Local
- Languages: TypeScript
- Best for: Testing MCP servers and OAuth flows in CI
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#mcpjam

### Metorial

Metorial is an MCP and skills gateway that puts integrations behind one MCP URL with SSO, role-based tool access, audit logs, and shared skills for Claude, ChatGPT, and Cursor; its code is source-available under the Functional Source License.

- Website: https://metorial.com
- GitHub: https://github.com/metorial/metorial
- Open source: No
- Deployment: Managed cloud, Self-hosted
- Best for: Company-wide MCP access behind single sign-on
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#metorial

### Pipedream

Pipedream is an integration layer for AI agents that provides managed authentication and prebuilt tools for thousands of APIs through Pipedream Connect and a remote MCP server, plus a gateway for the AI tools used inside a company.

Workday completed its acquisition of Pipedream, which it announced in November 2025.

- Company: Workday
- Website: https://pipedream.com
- Open source: No
- Deployment: Managed cloud
- Best for: Embedding user-authorized app tools in agent products
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#pipedream-mcp

### PulseMCP

PulseMCP is a directory and news site for the Model Context Protocol ecosystem that catalogs MCP servers, clients, and use cases and publishes updates on the protocol, and it contributes to the official MCP Registry.

- Website: https://www.pulsemcp.com
- Open source: No
- Deployment: Managed cloud
- Best for: Following MCP ecosystem changes
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#pulsemcp

### Smithery

Smithery is a public registry of MCP servers that lists verified servers with their vendors' direct MCP URLs and is now powered by Arcade.dev, which acquired Smithery in August 2026.

Smithery's MCP proxy, managed connections, API, accounts, and CLI shut down on October 31, 2026, while the public registry remains.

- Company: Arcade.dev
- Website: https://smithery.ai
- Open source: No
- Deployment: Managed cloud
- Best for: Looking up vendor MCP endpoints
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#smithery

### Speakeasy MCP Gateway

Speakeasy's MCP gateway is part of its AI control plane, giving agents and employees IdP-backed identity, a registry of approved MCP servers and tools, and a gateway that runs inside a company's own infrastructure, built on the open-source Gram stack.

Speakeasy's Gram product now redirects to its AI control plane, and the AGPL-3.0 Gram repository holds the open-source stack behind it.

- Company: Speakeasy
- Website: https://www.speakeasy.com/product/mcp-gateway
- GitHub: https://github.com/speakeasy-api/gram
- Open source: Yes (AGPL-3.0)
- Deployment: Managed cloud, Self-hosted
- Best for: Governed MCP access for employees and agents
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#speakeasy-mcp-gateway

### ToolHive

ToolHive is Stacklok's open-source MCP platform that runs each MCP server in an isolated container and enforces identity and access policy per request, with a Kubernetes-native enterprise edition adding a gateway, registry, and portal.

- Company: Stacklok
- Website: https://stacklok.com/platform/
- GitHub: https://github.com/stacklok/toolhive
- Open source: Yes (Apache-2.0)
- Deployment: Self-hosted, Local
- Languages: Go
- Best for: Platform teams running MCP servers on Kubernetes
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#toolhive

### Zapier MCP

Zapier MCP connects AI clients such as Claude, ChatGPT, and Cursor to actions in thousands of apps on Zapier's automation platform, with authentication and permissions managed by Zapier.

- Company: Zapier
- Website: https://zapier.com/mcp
- Open source: No
- Deployment: Managed cloud
- Best for: Giving assistants actions in many SaaS apps
- Permalink: https://theaiengineering.com/platforms/mcp-platforms#zapier-mcp

## How to choose

1. **Role**: Registries list server metadata, hosting platforms run servers, integration platforms supply ready-made authenticated tools, gateways govern access, and developer tools test servers.
2. **Provenance and security**: Namespace verification, container packaging with provenance, and security scanning reduce the risk of installing a malicious or abandoned server.
3. **Authentication**: Support for OAuth, per-user connected accounts, and token storage decides whether an agent can act on behalf of each end user.
4. **Deployment**: Managed hosting is quickest, while self-hosted, Kubernetes-native, or local options keep tool traffic inside your network.
5. **Governance**: Role-based tool access, approval policies, and audit logs matter once many employees and agents share the same servers.
6. **Openness**: Open-source registries, frameworks, and gateways can be self-hosted and extended, and compatibility with the official registry API eases moving between them.

## Frequently asked questions

### What is the MCP Registry?

The MCP Registry is the official, centralized catalog of metadata for publicly accessible Model Context Protocol servers, backed by contributors such as Anthropic, GitHub, PulseMCP, and Microsoft. It records where to find and how to install each server rather than hosting server code, and it is meant to be consumed by downstream marketplaces and private registries. The registry is in preview, so breaking changes may occur.

### What is an MCP gateway?

An MCP gateway is a proxy that sits between AI clients and many MCP servers, giving them one endpoint while enforcing authentication, role-based access to tools, and audit logging. Platform teams use gateways to control which servers employees and agents may reach. Bifrost, ToolHive, IBM's ContextForge, Metorial, and Speakeasy's MCP gateway are examples; Bifrost is also an AI gateway, so model calls and tool calls share one set of keys and logs.

### Where can I host an MCP server?

A remote MCP server can run on any platform that serves HTTP, and several make it simpler. Cloudflare runs MCP servers on Workers with OAuth support, Vercel deploys them as functions with the mcp-handler package, Prefect hosts FastMCP servers through FastMCP Gateway, and Manufact hosts servers built with mcp-use. Docker's MCP Toolkit runs servers locally as containers.

### Smithery vs Composio: what is the difference?

Smithery is a public registry of MCP servers; after Arcade.dev acquired it in August 2026, its hosted proxy, managed connections, API, and CLI shut down on October 31, 2026, and its registry pages now list vendors' own MCP URLs. Composio is an integration platform that provides authentication, tool discovery, and prebuilt toolkits for many apps through SDKs and MCP, so agents can act in users' accounts.

## Related guides

- [What is the Model Context Protocol (MCP)? How it works](https://theaiengineering.com/articles/what-is-model-context-protocol): The Model Context Protocol (MCP) is an open standard for connecting AI apps to tools and data. Learn its architecture, transports, tool calls, and security.
- [A2A protocol vs MCP: differences and when to use each](https://theaiengineering.com/articles/a2a-vs-mcp): A2A protocol vs MCP: A2A connects AI agents to other agents, and MCP connects an agent to tools and data. Compare agent cards, tasks, transports, and security.
- [What is tool calling? How LLM function calling works](https://theaiengineering.com/articles/tool-calling): Tool calling lets an LLM request actions from your code through structured calls. Learn the request loop, parallel calls, structured outputs, security, and MCP.

Product details change quickly; check each vendor's site for current features, licensing and pricing.

## Cite this page

The AI Engineering Editors. "MCP platforms." The AI Engineering. https://theaiengineering.com/platforms/mcp-platforms

- Title: MCP platforms
- Author: The AI Engineering Editors
- Publisher: The AI Engineering
- URL: https://theaiengineering.com/platforms/mcp-platforms
- Markdown: https://theaiengineering.com/platforms/mcp-platforms.md
