Developers

API and MCP server

Search the Agent Mart catalog from your code or your agent: a JSON API with an OpenAPI 3.1 description, and a remote MCP server agents can add in one line.

Use from your agent

The MCP server at https://agentmart.dev/mcp (streamable HTTP, no sign-in) gives an agent six read-only tools: search_listings, get_listing, get_install_config, compare_listings, list_categories and find_alternatives. Ask it, for example, “find a Postgres MCP server for Cursor and give me the config”.

Claude Code

claude mcp add --transport http agent-market https://agentmart.dev/mcp

Codex (~/.codex/config.toml)

[mcp_servers.agent-market]
url = "https://agentmart.dev/mcp"

Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "agent-market": {
      "url": "https://agentmart.dev/mcp"
    }
  }
}

VS Code (.vscode/mcp.json)

{
  "servers": {
    "agent-market": {
      "type": "http",
      "url": "https://agentmart.dev/mcp"
    }
  }
}

Listing names and descriptions are written by their publishers; the server marks them as data, and grades are automated assessments, not audits.

From the terminal

The agentmarket CLI searches the catalog and writes the right config for Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI and Claude Desktop.

npx agentmarket install connector/notion-com-mcp --host cursor

API quickstart

Base URL https://agentmart.dev/api/v1. JSON, CORS open for GET. OpenAPI 3.1: openapi.json.

curl 'https://agentmart.dev/api/v1/search?q=postgres&type=connector&limit=5'

const res = await fetch("https://agentmart.dev/api/v1/search?q=read+jira+tickets");
const { results, next_cursor } = await res.json();

import requests
r = requests.get("https://agentmart.dev/api/v1/listings/connectors/notion-com-mcp/install", params={"host": "claude-code"})
print(r.json()["options"][0]["command"])

Keys and limits

Tier	Limit	How
No key	60 per hour and 500 per day per IP	Just call it
Free key	60 per minute, 10,000 per day	Authorization: Bearer am_live_…
Partner	Higher, by agreement	Contact us

Every response carries X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset; over the limit you get 429 with Retry-After. Anonymous responses are cached for five minutes. Self-serve keys arrive with accounts; until then, ask for a key. Errors look like {"error": {"code", "message", "docs_url"}}.

Endpoints

GET /search

Search listings

q (query) Search text. Typos, synonyms and words like "mcp" or "skill" are understood.
type (query: connector, skill, plugin, harness, agent, builder, model, infrastructure, …) Listing type
host (query: claude-code, codex, cursor, vscode, windsurf, claude-desktop, gemini-cli, github-copilot, …) Only listings that work with this host
transport (query: streamable-http, sse, stdio)
auth (query: none, api-key, oauth, bearer, unknown)
license (query: open-source)
official (query) true: only listings from the product's own publisher
include_low (query) true: include listings below the quality bar
read_only (query) true: only listings whose tools are all known and only read (none write, delete or are unclassified)
live (query) true: only remote connectors that answered their last daily check (up, or asking for auth)
no_secrets (query) true: only listings whose permission manifest asks for no API keys, tokens or passwords
sort (query: relevance, rank, newest, stars, downloads, updated, name)
limit (query)
cursor (query) Opaque cursor from the previous page's next_cursor

GET /listings/{type}/{slug}

One listing

type (path, required) Type slug as in the site URL, e.g. connectors, skills, plugins
slug (path, required) Listing slug
fields (query) Comma-separated top-level fields to return, e.g. id,name,grades
expand (query) Comma-separated: tools, relations, sources

GET /listings/{type}/{slug}/install

Install command or config for one host

type (path, required) Type slug as in the site URL, e.g. connectors, skills, plugins
slug (path, required) Listing slug
host (query, required: claude-code, codex, cursor, vscode, windsurf, claude-desktop, gemini-cli, github-copilot, …) Host to install into

GET /listings/{type}/{slug}/alternatives

Similar listings of the same type

type (path, required) Type slug as in the site URL, e.g. connectors, skills, plugins
slug (path, required) Listing slug
limit (query)

GET /compare

Compare 2 to 4 listings

ids (query, required) 2 to 4 listing ids, comma-separated (connector:io.github.foo/bar)

GET /types

Listing types with counts

GET /taxonomy

Ids accepted by the search filters

GET /publishers/{slug}

A publisher and its top listings

slug (path, required) gh-<org> or dns-<domain>

GET /listings/{type}/{slug}/reviews

Published reviews of a listing

type (path, required) Type slug as in the site URL, e.g. connectors, skills, plugins
slug (path, required) Listing slug
host (query)
version (query)
limit (query)
cursor (query)

GET /stacks

Public stacks

use_case (query) Use case id, e.g. coding, data, research, support
host (query: claude-code, codex, cursor, vscode, windsurf, gemini-cli, claude-desktop) Only stacks that install fully into this host
sort (query: installs, newest)
limit (query)
cursor (query) Opaque cursor from the previous page's next_cursor

GET /stacks/{slug}

One stack and its items

slug (path, required) Stack slug

GET /stacks/{slug}/install-plan

What installing the stack into one host does

slug (path, required) Stack slug
host (query, required: claude-code, codex, cursor, vscode, windsurf, gemini-cli, claude-desktop) Host to install the stack into

GET /stats

Catalog size and last sync

Versions and changes

v1 field names are fixed; new fields may appear, nothing is removed without 90 days' notice. v0 (docs) keeps working and is deprecated.

2026-11-12: v1 and the MCP server.

Terms (draft)

Show “Data from Agent Mart” with a link to https://agentmart.dev/ wherever you display results.
No resale or bulk redistribution of the catalog without a license; there is no bulk export in v1.
Stay within the limits. Keys that hit their limit continuously for three days are paused.
We may change limits or revoke keys used for abuse. Grades are automated assessments, not security audits or endorsements.