Attach once.
Use local stdio for a desktop or coding agent, or run the Streamable HTTP transport behind your own authenticated infrastructure.
Connect an MCP-compatible agent to one searchable index of reusable capabilities. Search by the work you need done, inspect the source, and decide what deserves a place in your workflow.
The MCP connection makes the directory available where the work happens. Your agent can find a capability without you having to leave the conversation or memorize another registry.
Use local stdio for a desktop or coding agent, or run the Streamable HTTP transport behind your own authenticated infrastructure.
Search by task, topic, source, or trust signal. The gateway can combine the local snapshot with live Skills.sh results.
Read metadata, install paths, provenance, and an explicitly requested skill file as untrusted text before making a decision.
EveryAI does not silently install or execute third-party code. The agent and its operator decide what enters the environment.
The local server runs from the EveryAI repository and speaks standard MCP over stdio. Point your agent at the file, then search from the client you already use.
npm install
npm run mcp
Run those commands from the EveryAI project directory.
{
"mcpServers": {
"everyai-skills": {
"command": "node",
"args": [
"/absolute/path/to/everyai/mcp/server.mjs"
]
}
}
}
Use the absolute path to mcp/server.mjs in your agent's MCP configuration.
Need the full configuration notes? Read the implementation guide ↗ or learn about MCP servers ↗.
Each tool does one understandable job. The contract is intentionally small so an agent can use it quickly and an operator can audit it easily.
Find capabilities by query, topic, official status, install history, or live registry results.
Return one record with its source, description, install command, topics, and trust metadata.
Explicitly fetch a catalogued GitHub SKILL.md for inspection as untrusted text.
See the registries and publisher repositories that feed the EveryAI metadata layer.
The gateway supports Streamable HTTP for teams that need a shared endpoint. Keep it private by default: use TLS, an authentication token, and a rate-limiting reverse proxy before binding beyond localhost.
MCP_TRANSPORT=http \
MCP_HOST=127.0.0.1 \
MCP_PORT=8788 \
MCP_AUTH_TOKEN=replace-me \
npm run mcp:http
The server refuses non-local binding when MCP_AUTH_TOKEN is missing.
It lets an MCP-compatible agent search the EveryAI skills index, inspect metadata and provenance, list connected registries, and explicitly read a catalogued SKILL.md as untrusted text.
No. The gateway is metadata-first and read-only by default. It does not silently install or execute third-party code.
Yes. Use the Streamable HTTP transport behind TLS and a rate-limiting reverse proxy, and set MCP_AUTH_TOKEN before binding to a non-local interface.
Any MCP-compatible agent or client can use the server. Local stdio is the simplest option for desktop and coding agents; HTTP is available for self-hosted setups.
EveryAI indexes public metadata and keeps publisher links visible. Read the MCP tools specification ↗ and the Agent Skills specification ↗ for the underlying standards.