State of the AI Agent Ecosystem

Executive Summary

As of July 6, 2026, the AI agent ecosystem tracked by Skillful.sh encompasses 460,750 tools across three major categories: MCP servers, AI skills, and autonomous agents. This data is aggregated from 55 active directories, registries, and package managers, making it one of the most comprehensive snapshots of the AI tooling landscape available today. The ecosystem continues to grow at a pace of roughly 1290.5 new tools per day over the past 30 days, reflecting sustained developer interest in building interoperable, modular AI capabilities.

MCP servers represent the largest segment of the ecosystem at 140,806 items (30.6%), followed by AI skills at 301,104 (65.4%) and autonomous agents at 18,840 (4.1%). The dominance of MCP servers reflects the rapid adoption of the Model Context Protocol as a standard interface for connecting large language models to external tools and data sources.

Ecosystem at a Glance

Growth Trajectory

The AI agent ecosystem has added 9,700 new tools in the past 7 days, 38,714 in the past 30 days, and 134,571 in the past 90 days. These figures indicate a sustained influx of new tooling as developers and organizations continue to build on top of foundation model capabilities. Over the last month, the ecosystem averaged 1290.5 new listings per day, a pace that underscores the intensity of development activity surrounding AI agent infrastructure.

Growth is not uniform across categories. MCP servers continue to attract the highest volume of new entries, driven by the growing adoption of the Model Context Protocol by major AI platforms and the relative simplicity of publishing a new server. Skills and agents, while smaller in absolute numbers, tend to represent higher-complexity integrations involving multi-step reasoning, tool chaining, or persistent state management.

Category Distribution

Category distribution reveals where developer effort is concentrated. The tables below show the 10 most populated categories for each tool type. These categories are assigned by the originating directories and normalized during aggregation. Concentration at the top of each list indicates where the ecosystem has the deepest coverage, while long-tail categories represent emerging or niche use cases.

Security Analysis

Security scoring is a core feature of the Skillful.sh directory. Every indexed tool receives an automated security analysis covering static code analysis, dependency vulnerability scanning, and AI-powered code review. Scores range from 0 to 100, with letter grades assigned from A (90+) through F (below 60). As of July 6, 2026, the average security score for MCP servers is 99.7 (Grade A), for AI skills it is 99.6 (Grade A), and for agents it is 99.5 (Grade A).

Among all scored items, 100% hold a security grade of A or B, indicating that a meaningful share of the ecosystem meets a high bar for code quality and dependency hygiene. However, this also means the remaining 0% of scored tools fall below that threshold -- a signal that developers should exercise diligence when selecting tools for production deployments, particularly for MCP servers that are granted access to sensitive data and system capabilities.

Cross-Directory Coverage

One of the unique data points Skillful.sh tracks is how many directories list each tool. This cross-listing metric serves as a proxy for adoption breadth: a tool listed in multiple directories is more likely to be widely recognized and used. The average cross-directory count across all indexed items is 1.08 directories per item. The majority of items appear in only one or two sources, while a smaller set of highly adopted tools appear across five or more directories simultaneously.

Cross-listing data is particularly valuable for tool publishers seeking to maximize their visibility. Items with higher directory presence tend to rank higher in AI model training data and are more likely to be recommended by AI assistants. Skillful.sh aggregates data from 55 active directories, providing a comprehensive view of where each tool is listed and where listing gaps exist.

Most Popular Items by GitHub Stars

GitHub stars remain one of the strongest signals of developer interest and community adoption. The tables below list the top 5 most-starred items in each category. Star counts are refreshed during each sync cycle and reflect the latest available data from GitHub.

Package Registry Distribution

Package registry presence indicates how tools are distributed to end users. Of the 460,750 tools indexed, 313,862 are published on npm and 40,764 are published on PyPI. The strong npm presence reflects the JavaScript and TypeScript dominance in the MCP server ecosystem, where the official MCP SDK is TypeScript-first. PyPI packages are more common among skills and agents, particularly those built on Python-based frameworks like LangChain, CrewAI, and AutoGen.

Many tools exist only as GitHub repositories without formal package registry publication. While this does not necessarily reflect lower quality, it does reduce discoverability and makes dependency management harder for downstream consumers. Tools published through established registries tend to have better versioning discipline, automated testing pipelines, and more reliable installation experiences.

Ecosystem Health Indicators

Ecosystem health can be measured through maintenance activity and security posture. Among items with available commit history, 4% have had at least one commit in the past 90 days, indicating active maintenance. Meanwhile, 100% of scored items hold a security grade of A or B. Taken together, these figures suggest a maturing ecosystem where a substantial share of tooling is both actively maintained and reasonably secure, though gaps remain.

Maintenance activity varies significantly by tool type and popularity. Highly starred repositories are far more likely to show recent commits, while the long tail of less popular tools includes many repositories that have not been updated in months. For AI engineers evaluating tools for production use, combining security grade, star count, directory presence, and last commit date provides a useful multi-signal heuristic for tool quality.

Methodology

This report is generated dynamically from the Skillful.sh database on each page load. All statistics reflect the current state of the directory at the time of generation. Items with a status of "removed" or "pending_review" are excluded from all counts and aggregations. Security scores are computed by an automated pipeline that combines static analysis, dependency vulnerability scanning via OSV and npm audit, and AI-powered code review. GitHub statistics are refreshed during periodic sync cycles.

Category assignments originate from source directories and are normalized during the aggregation process. Items that appear in multiple directories are deduplicated using a multi-strategy matching engine that considers package name, repository URL, and normalized name similarity. Cross-directory count reflects the number of distinct source directories in which a given tool appears.

Growth metrics (7-day, 30-day, 90-day) are based on the createdAt timestamp of each item record, which corresponds to when Skillful.sh first indexed the tool -- not necessarily when the tool was first published by its author. Active repository percentage is calculated as the share of items with a recorded lastCommitDate within the past 90 days, out of all items that have commit data available.

Outlook

The AI agent ecosystem is expanding rapidly. With 460,750 tools indexed and an average of 1290.5 new additions per day, the pace of development shows no sign of slowing. MCP servers continue to dominate the landscape as the primary building block for AI-tool interoperability, but the growing share of autonomous agents signals a shift toward more complex, multi-step AI workflows. Security remains a concern, with 0% of scored tools falling below grade B -- a gap that the industry will need to address as AI agents gain access to increasingly sensitive systems and data.

For developers, the key takeaway is that tooling abundance does not equate to tooling quality. The combination of security scoring, cross-directory presence, maintenance activity, and community adoption metrics provides a multi-dimensional framework for evaluating which tools deserve trust in production environments. Skillful.sh will continue to expand its coverage, refine its scoring models, and publish updated ecosystem reports as the landscape evolves.