From: aidotengineer
The Model Context Protocol (MCP) is a relatively new space [00:00:02]. This article provides a high-level overview and addresses the various problems within the MCP space and ecosystem [00:00:14]. Henry, the founder and CEO of Smithery and a member of the MCP committee, shares insights into these challenges [00:00:19].
The Genesis of MCP
The need for MCP emerged from a realization regarding the capabilities of Large Language Models (LLMs). While LLMs, such as OpenAI’s 03 (which achieved human-level performance on challenges like ARC AGI) [00:01:15], demonstrated significant intelligence, this intelligence often remained “stuck in a box,” a phenomenon referred to as “Claude’s paradox” [00:01:45]. To make AI agents practically useful, there was a need to consider their context and capability, specifically their inputs and outputs [00:01:56].
Recognizing this problem, Anthropic released the Model Context Protocol in November 2024 [00:02:08]. This open standard aims to help LLMs connect to different services, standardizing the complex problem of connecting models to external functionalities [00:02:11]. The advent of MCP and a new ecosystem of services targeting AI agents, while promising, also introduced a new set of problems [00:02:34].
User Problems in the MCP Ecosystem
Users of MCPs face several significant challenges:
- Fragmentation: With an increasing number of MCP servers being deployed daily, it is becoming difficult to find high-quality ones [00:02:50]. The MCP committee is working on an official registry [00:02:56], but assigning reputation to high-quality MCPs remains an open question [00:03:07].
- High Friction Installation: Installing an MCP typically involves a complex five-step process found in their GitHub repositories, making it very difficult [00:03:14]. Furthermore, there’s a risk of installing an insecure MCP [00:03:24].
- AI Native Services Economy: The creation of a new economy for AI native services faces challenges, particularly regarding agentic payments [00:03:30]. Questions include how agents will pay on behalf of users and how to prevent users from subscribing to numerous services each charging recurring fees [00:03:37].
Developer Problems in the MCP Ecosystem
Developers building MCPs also encounter a distinct set of challenges:
- Hosting: Although Streambo HTTP transport has made hosting easier [00:03:54], developers still contend with issues like stable sessions and resumability [00:04:01].
- Developer Tooling: The MCP space currently lacks robust developer tooling, with only a basic MCP inspector available from the official MCP repository [00:04:06]. Open questions for developers include how to design the best MCPs, ensure their tools are called, and create optimal agent experiences [00:04:21].
- Distribution: Developers face the challenge of getting their created MCPs discovered [00:04:30].
- Observability: Once an MCP is deployed and in use, developers need ways to monitor and improve it [00:04:36].
- Monetization: A key challenge is determining how developers can generate revenue from their MCPs [00:04:43].
Smithery’s Approach to Solving MCP Challenges
Smithery was founded in December 2024 to tackle these ecosystem challenges [00:04:58]. Its goal is to become the “AI gateway” that grows and orchestrates the new era of AI native services for AI agents [00:05:03].
Demonstration of Smithery’s Capabilities
Smithery’s playground demonstrates what an AI agent can achieve with access to thousands of curated MCPs [00:05:22]. For example, an agent can be prompted to “Find the most pressing issue on my GitHub repository called smidy-CLI and create a new ticket on Linear” [00:05:31].
The agent’s process involves:
- Thinking about the issue [00:05:48].
- Calling a search servers function to find relevant servers within Smithery [00:05:51].
- Connecting to the best server, including Linear [00:05:55].
- Using the GitHub MCP to find the most high-priority bug [00:06:06].
- Creating a ticket on Linear with details and a link to the original issue [00:06:15].
This end-to-end task showcases an AI agent connected to two different MCPs [00:06:40], demonstrating the potential when MCP problems are addressed.
The Future of the Internet and AI Agents
The rapid increase in MCP server deployments and tool calls [00:06:55] indicates a strong developer enthusiasm for the MCP ecosystem [00:07:01]. It is becoming clear that the future of the internet will be dominated by “tool calls” rather than traditional “clicks” [00:07:09]. In this evolving landscape, the “agent experience” will supersede the “user experience” in importance [00:07:15], and this future will be built collaboratively by the entire community [00:07:23].