Structured data from 106 platforms (RedNote, Douyin, TikTok, Amazon), web search, CAPTCHA solving.
MCP Server: io.github.xiaoxihexiaoyu/everyinfra
This MCP server provides APIs and MCP access for structured public data, web research, CAPTCHA, and source-bound data cleanup. It is identified by the slug io.github.xiaoxihexiaoyu/everyinfra and exposes 8 tools for these data-handling tasks.
🛠️ Key Features
APIs and MCP integration
Structured public data handling
Web research
CAPTCHA support
Source-bound data cleanup
🚀 Use Cases
Retrieving and structuring public data via MCP
Performing web research workflows
Handling CAPTCHA as part of data collection
Cleaning or normalizing data with cleanup tied to original sources
⚡ Developer Benefits
Centralized MCP + API surface for the listed capabilities
Tool-oriented interface (8 tools) aligned to structured data, research, CAPTCHA, and cleanup needs
⚠️ Limitations
Limited scope as described to the four listed domains (structured public data, web research, CAPTCHA, and source-bound cleanup)
List EveryInfra's data platforms and capabilities (106 platforms / 573 capabilities: search, detail, comments, profiles and more). Use it before everyinfra_call_api to get the platform/action, required and optional params, the per-call row limit (max_limit, or page_size for page-based actions) and the price. Pass a platform to also get allowed values, returned fields and what they mean. Free. Data collection only — for AI chat use everyinfra_chat.
Parameters1
platform
string
optional
Optional; only this platform's capabilities, e.g. xiaohongshu / twitter / tiktok
Raw schema
{
"type": "object",
"properties": {
"platform": {
"type": "string",
"description": "Optional; only this platform's capabilities, e.g. xiaohongshu / twitter / tiktok"
}
}
}
everyinfra_call_api
Call any EveryInfra data capability and get structured JSON. Priced per platform tier USD 0.56/1K–USD 5.56/1K; failed calls and empty results leave no final charge (anything reserved is restored). Most actions return up to `limit` rows in one call, billed once; multi-target arrays (urls, targets, usernames, tickers) are billed per element, and actions with rows_per_unit per block of rows asked for. Page-based actions (page_size in the catalog) return one page per call, of a fixed size unless page_size_varies (then a short page is not the end); for more, pass the response's top-level next_page_token (some actions repeat it on each row) as page_token — each page is one call. Check platform/action, required params and max_limit or page_size with everyinfra_list_capabilities first. For slow tasks or large limits pass mode=async: you get a job_id at once; read the result with everyinfra_get_job (free).
Parameters4
platform
string
required
e.g. xiaohongshu / twitter / tiktok / web
action
string
required
e.g. search / profile / user_posts / post
params
object
optional
The capability's params, e.g. {"keyword":"AI tools","limit":20}
mode
string
optional
sync (default) waits for the result; async returns a job_id at once for slow tasks or large limits — read it with everyinfra_get_job
Raw schema
{
"type": "object",
"properties": {
"platform": {
"type": "string",
"description": "e.g. xiaohongshu / twitter / tiktok / web"
},
"action": {
"type": "string",
"description": "e.g. search / profile / user_posts / post"
},
"params": {
"type": "object",
"description": "The capability's params, e.g. {\"keyword\":\"AI tools\",\"limit\":20}"
},
"mode": {
"type": "string",
"enum": [
"sync",
"async"
],
"default": "sync",
"description": "sync (default) waits for the result; async returns a job_id at once for slow tasks or large limits — read it with everyinfra_get_job"
}
},
"required": [
"platform",
"action"
]
}
everyinfra_get_job
Read a job submitted with everyinfra_call_api(mode=async). Free; it does not count against the per-minute limit. While status is submitting / running / finalizing, check again later; succeeded: the rows are in results (paged actions also return next_page_token); failed: see error.reason and error.charged; submit_unknown: the submission is still being confirmed — do not resubmit, follow next_action. Only jobs submitted with the same API key are visible.
Parameters1
job_id
string
required
The job_id returned by an async everyinfra_call_api
Raw schema
{
"type": "object",
"properties": {
"job_id": {
"type": "string",
"description": "The job_id returned by an async everyinfra_call_api"
}
},
"required": [
"job_id"
]
}
everyinfra_chat
EveryInfra's OpenAI-compatible text completions with multi-turn history and any input content or format. Free for accounts that have topped up (nothing is deducted); accounts that have not get ai_topup_required. Rate-limited per account by cumulative top-ups (under $100: 50 requests/minute, $100 or more: 500, $500 or more: 5,000, all keys combined); current models: GET /api/v1/models. Good for cleaning collected data: summaries, translation, sentiment labels, topic classification and structured extraction. messages may total up to 45,000 characters; beyond that you get input_too_large — split long documents into several calls.
Parameters2
messages
array
required
OpenAI-format conversation, e.g. [{"role":"user","content":"Hello"}] (role is system / user / assistant).
List the 52 CAPTCHA and bot-challenge types EveryInfra solves. **Call it before everyinfra_solve_captcha** to get three things: the type name, the required/optional params, and **the shape of the solution** (token / cookie / text / number / fields / points / boxes / tokens / indices — a string to submit, coordinates to click yourself, or grid cells to click: each is used differently). Free.
Parameters1
type
string
optional
Optional; only this type, e.g. turnstile / recaptcha_v2 / cloudflare_challenge
Raw schema
{
"type": "object",
"properties": {
"type": {
"type": "string",
"description": "Optional; only this type, e.g. turnstile / recaptcha_v2 / cloudflare_challenge"
}
}
}
everyinfra_solve_captcha
Solve a CAPTCHA or bot challenge and get a ready-to-use solution. USD 0.000139/call–0.004583/call, **charged per successful solve (castle: per token, × count); failed or empty attempts leave no final charge (anything reserved is restored)**. Use it when a target site is blocked by Turnstile / reCAPTCHA / hCaptcha / a Cloudflare challenge page / GeeTest and the like. Params differ per type — check them with everyinfra_list_captcha_types first.
Parameters2
type
string
required
CAPTCHA type, e.g. turnstile / recaptcha_v2 / image
params
object
optional
Params for that type, e.g. {"website_url":"https://…","website_key":"0x…"}. ⚠ A few types need your own proxy: they deliver a cookie bound to the IP the challenge was solved from, so a solution from our exit IP would not work for you
Raw schema
{
"type": "object",
"properties": {
"type": {
"type": "string",
"enum": [
"akamai_bmp",
"akamai_sec_cpt",
"akamai_web",
"alibaba",
"altcha",
"anubis",
"atb_captcha",
"audio",
"aws_waf",
"aws_waf_classify",
"basilisk",
"binance",
"botdeflector",
"bounding_box",
"captchafox",
"capy",
"castle",
"cloudflare_challenge",
"coordinates",
"cutcaptcha",
"cybersiara",
"datadome",
"datadome_interstitial",
"f5_shape",
"f5_tspd",
"friendly_captcha",
"funcaptcha",
"geetest",
"hcaptcha",
"hunt",
"image",
"imperva",
"kasada",
"kasada_cd",
"lemin",
"mtcaptcha",
"perimeterx",
"prosopo",
"recaptcha_grid",
"recaptcha_v2",
"recaptcha_v2_enterprise",
"recaptcha_v3",
"recaptcha_v3_enterprise",
"rotate",
"shein",
"slider",
"temu",
"tencent_tendi",
"text_captcha",
"turnstile",
"yandex_smartcaptcha",
"yidun"
],
"description": "CAPTCHA type, e.g. turnstile / recaptcha_v2 / image"
},
"params": {
"type": "object",
"description": "Params for that type, e.g. {\"website_url\":\"https://…\",\"website_key\":\"0x…\"}. ⚠ A few types need your own proxy: they deliver a cookie bound to the IP the challenge was solved from, so a solution from our exit IP would not work for you"
}
},
"required": [
"type"
]
}
everyinfra_search
Search and retrieve sources. 17 tools, chosen with `tool`:
crawl: Crawl several pages of a site and return their text; say in plain language which pages you want.
crosscheck: Run one question through several retrieval chains with different mechanisms, rank sources by how many mechanisms found them, and fetch the original text of the top tier. It answers how far a claim can be trusted, not just what was found.
deep: Deep search with text excerpts; full=true also returns the full text.
forum: Real people's discussions — an evidence source independent of official docs (when docs say 'supported' and forums describe where it breaks, trust the specific scenario).
harvest: Structured site extraction (long-running: tens of seconds to minutes); use it only when crawl cannot get the data.
lens: Reverse image search: give an image URL, find pages that use it (the opposite direction of media).
map: List a site's page structure — URLs only, no text, fast.
media: Find images and videos by keyword.
news: News search.
places: Places and businesses search.
read: Fetch the text of a given URL; it retries other fetch paths automatically when one fails.
scholar: Academic literature search; kind=patents switches to patents (publication number, applicant, inventor, filing and publication dates).
semantic: Semantic search: describe what you want in one sentence instead of stacking keywords; it finds deep sources whose titles do not contain your words. mode=deep searches deeper (slower, costs more).
shopping: Product search and price comparison.
similar: Give a URL, find semantically similar pages — something keyword search cannot do.
suggest: Search suggestions (autocomplete).
web: General web search with site restriction, region, time window and paging.
⚠ To **assert a specific fact** (a version, a price, whether an API exists), use crosscheck: it returns verifiable sources and their agreement tier. Failed or empty searches leave no final charge (anything reserved is restored).
Parameters2
tool
string
required
Retrieval tool
params
object
optional
The tool's params, e.g. {"q":"EU AI Act","num":10}
Read the account behind this API key. Free, read-only, any key works. Returns:
- balance: total_usd, paid_usd, bonus_usd and bonus_expiring (amount_usd, expires_at)
- discount: this month's top-ups (month_topup_usd), the rate in effect and next (rate, threshold_usd, gap_usd = how much more to top up this month for it)
- rate_limit: requests_per_minute for all keys combined, key_requests_per_minute, and any rate package with its expires_at
- key: this key's id, name, prefix, created_at and expires_at
- usage_24h: calls, succeeded, failed and spent_usd in the last 24 hours (this_key: the same for this key)
Check the balance before a batch of paid calls; every paid call's billing also carries balance_after_usd.
Parameters
No parameters.
Raw schema
{
"type": "object",
"properties": {}
}
everyinfra_list_requests
The account's call records, newest first: request_id, capability, status code, latency, rows returned, whether it was charged, and the amount. Pass request_id (the request_id / X-Request-Id of any call) to look up one call, including its charge's final state (charged, refunded or partly refunded) and any async job it started. failed_only=true lists failures only. Needs the account read permission "requests:read" on this key: keys not limited to product lines have the three read permissions by default; for a key limited to product lines, the account owner turns it on in the console (API Keys → permissions). Free.
Parameters4
request_id
string
optional
Look up one call; the other params are ignored when given
failed_only
boolean
optional
List failed calls only, default false
limit
integer
optional
Rows, default 20
offset
integer
optional
Rows to skip, default 0
Raw schema
{
"type": "object",
"properties": {
"request_id": {
"type": "string",
"description": "Look up one call; the other params are ignored when given"
},
"failed_only": {
"type": "boolean",
"description": "List failed calls only, default false"
},
"limit": {
"type": "integer",
"minimum": 1,
"maximum": 50,
"description": "Rows, default 20"
},
"offset": {
"type": "integer",
"minimum": 0,
"maximum": 1000,
"description": "Rows to skip, default 0"
}
}
}
everyinfra_get_topup_link
Returns the console top-up page link (url) and the current balance (balance_usd). Use it when a call fails with quota_exhausted or ai_topup_required, or when the user asks how to add funds. The account owner pays in the console; this tool creates no order and charges nothing. Any key can use it; accounts whose service provider handles top-ups get topup_unavailable. Free.
Parameters
No parameters.
Raw schema
{
"type": "object",
"properties": {}
}
everyinfra_create_ticket
Open a ticket with EveryInfra's human support team and get back the ticket number and a link the account owner can open. Use it when calls keep failing or return wrong or missing data, for top-up, charge or refund questions, or to ask for a new platform, capability or field. Put one line in subject; in description say what you called (platform/action or endpoint), when (UTC), the error code and message and what you expected, and pass that call's request_id (and job_id for an async job); write other IDs with a label such as job_id: job_..., because a bare long ID reads as a secret. Never include API keys or passwords. Ask the user first and open it only if they agree. Errors whose fault is request or account need a fixed request or the user, not a ticket; an error's support field carries a ready draft. Any key on the account works and nothing is charged. Retrying the same request returns the same ticket, and the same capability and error code within 24 hours goes into that ticket. Support is online 10:00-24:00 (UTC+8) with a 10-minute average first reply; replies arrive in the ticket and the account owner also gets an email. Check progress and read support's replies with everyinfra_get_ticket (REST: GET /api/v1/support/tickets/{ticket_id}) using the same key, or at the returned link.
Parameters7
subject
string
required
One line, e.g. "douyin/comments returns 503 since 10:20 UTC"
description
string
required
What you called, when (UTC), the error code and message, what you expected; no API keys or passwords
category
string
optional
Optional: incident (looks like an outage), integration (a call returns an error; parameters, auth, SDK or MCP setup), product (questions about data, platforms and fields, or asking for new ones), billing (top-ups, charges, refunds), order (resources such as proxies or numbers), general
request_id
string
optional
The request_id (X-Request-Id) of the call this ticket is about
job_id
string
optional
The job_id of the async job this ticket is about
contact_email
string
optional
Optional extra address for support to reach you; reply emails still go to the account's verified email
idempotency_key
string
optional
Optional; keep it when retrying. Without it, the same subject and description sent again within 24 hours return the same ticket
Raw schema
{
"type": "object",
"properties": {
"subject": {
"type": "string",
"maxLength": 160,
"description": "One line, e.g. \"douyin/comments returns 503 since 10:20 UTC\""
},
"description": {
"type": "string",
"maxLength": 3500,
"description": "What you called, when (UTC), the error code and message, what you expected; no API keys or passwords"
},
"category": {
"type": "string",
"enum": [
"general",
"product",
"integration",
"billing",
"order",
"incident"
],
"description": "Optional: incident (looks like an outage), integration (a call returns an error; parameters, auth, SDK or MCP setup), product (questions about data, platforms and fields, or asking for new ones), billing (top-ups, charges, refunds), order (resources such as proxies or numbers), general"
},
"request_id": {
"type": "string",
"pattern": "^req_[a-f0-9]{16,64}$",
"description": "The request_id (X-Request-Id) of the call this ticket is about"
},
"job_id": {
"type": "string",
"pattern": "^job_[a-f0-9]{16,64}$",
"description": "The job_id of the async job this ticket is about"
},
"contact_email": {
"type": "string",
"maxLength": 254,
"description": "Optional extra address for support to reach you; reply emails still go to the account's verified email"
},
"idempotency_key": {
"type": "string",
"maxLength": 255,
"description": "Optional; keep it when retrying. Without it, the same subject and description sent again within 24 hours return the same ticket"
}
},
"required": [
"subject",
"description"
]
}
everyinfra_get_ticket
Read one support ticket: its status, what happens next, and its latest 100 messages, including support's replies. Use it after everyinfra_create_ticket to check progress, or when the user asks whether support has answered. Pass ticket_id as a plain integer (the number shown as #N, without the #). The key that opened a ticket can always read it; reading the account's other tickets (opened in the console or with other keys) needs the account read permission "support:read" on this key. Any other number returns ticket_not_found. Free; the same result as GET /api/v1/support/tickets/{ticket_id}.
Parameters1
ticket_id
integer
required
The ticket_id everyinfra_create_ticket returned (#N in the console and emails)
Raw schema
{
"type": "object",
"properties": {
"ticket_id": {
"type": "integer",
"minimum": 1,
"description": "The ticket_id everyinfra_create_ticket returned (#N in the console and emails)"
}
},
"required": [
"ticket_id"
]
}
One API key and one MCP endpoint for public web data: 573 data APIs across 106 platforms, 17 web search tools and CAPTCHA solving for 52 types. From $0.56 per 1,000 calls; failed or empty calls are not charged.
Without EveryInfra
With EveryInfra
❌ Your agent can't read TikTok, Instagram or Xiaohongshu: logins and blocked requests stop it.
❌ One scraper per platform, rewritten every time the page changes.
❌ You pay for runs that return nothing.
✅ Your agent picks the API and checks the inputs itself.
✅ Every platform comes back in the same JSON shape.
Numbers come from the live catalog on 2026-10-24. The live catalog and MCP tool schemas are authoritative.
What is in this repository
This repository packages the EveryInfra agent plugin and the remote MCP connection metadata. It is
not a copy of the EveryInfra backend. It contains:
11 discoverable skills: one router, seven product skills, two outcome workflows, and the
everyinfra-cli skill for the command-line tool;
remote MCP connection metadata for https://api.everyinfra.com/mcp;
host-specific plugin manifests for Codex, Claude Code, Cursor, Qoder, CodeBuddy, and Kimi;
self-hosted marketplace manifests for Codex, Claude Code, Cursor, GitHub Copilot, Qoder, and
CodeBuddy;
package sources for markets that take uploads ( MiniMax, Qoder/ QoderWork, ClawHub) under
host-packages/;
presentation assets and safety boundaries for paid calls, credentials, billing, and external side
effects.
No API key, customer data, credential, or private gateway source code belongs in this repository.
MCP and plugins are different layers. MCP is the runtime protocol for live tools, resources,
prompts, and server instructions. A plugin is an installable package that can bundle skills, MCP
connection metadata, hooks, UI assets, and host-specific discovery metadata. Either layer can be
used without the other. Many hosts show the installed package through its skills: seeing
everyinfra, everysearch, or the other child skills in a skill list is expected, and it does not
by itself prove that the MCP connection is authenticated. See
MCP, plugin, app, and marketplace boundaries.
What agents can use it for
"Find current sources for this claim and cross-check them."
"Discover the right public-data capability, then export a bounded CSV."
"Search public profiles, posts, comments, listings, or reviews as structured data."
"Clean and label these collected comments with the AI API, and return JSON."
"Solve this captcha for a target I am authorized to automate."
"Inspect available phone-number, transactional-email, or proxy options before I approve an
order or send."
AI discovery terms include: EveryInfra, EveryData, EverySearch, EveryAI, EverySolve,
EveryNumber, EveryMail, EveryProxy, MCP, web search, cited research, structured public data,
social data, ecommerce data, bulk export, captcha, phone verification, transactional email, and
proxy API.
For an AI agent:
Use the everyinfra router when the user names EveryInfra, needs more than one product, or has
not chosen a product.
Use a product skill directly when the requested outcome is unambiguous.
Discover live MCP capabilities and schemas before inventing parameters or estimating cost.
Treat purchases, sends, rentals, captcha solves, and credential delivery as explicit external
actions. A read or draft request is not authorization to execute them.
Keep EveryNumber, EveryMail, and EveryProxy on their documented REST paths until live MCP
discovery proves otherwise.
Capability matrix
Product or workflow
Package entry
Runtime path
Typical outcome
EveryInfra
everyinfra
Router
Choose the correct product and execution boundary.
EveryData
everydata
MCP
Discover and retrieve structured public-platform data.
EverySearch
everysearch
MCP
Search, read, crawl, and cross-check current public sources.
EveryAI
everyai
MCP
Run OpenAI-compatible text generation through EveryInfra.
EverySolve
everysolve
MCP
Handle an authorized captcha or anti-bot challenge.
EveryNumber
everynumber
REST guidance
Inspect or operate authorized number activation and rental flows.
EveryMail
everymail
REST guidance
Inspect, draft, schedule, or send transactional email with authorization.
EveryProxy
everyproxy
REST guidance
Inspect packages, order, and retrieve credentials with authorization.
Cross-check research
everyinfra-crosscheck-research
MCP workflow
Produce an evidence-backed answer from distinct retrieval paths.
Bulk data export
everyinfra-bulk-data-export
MCP workflow
Export bounded, deduplicated CSV or JSON with cost controls.
Command line
everyinfra-cli
CLI
Run data, search, CAPTCHA and AI calls from a shell with a dry-run cost preview and a budget cap.
The live capability catalog and tool schemas are authoritative. The package must not invent an MCP
tool name, current price, availability state, or billing result.
AI API route. Use the MCP tool everyinfra_chat, or REST
POST https://api.everyinfra.com/api/v1/chat/completions with any OpenAI SDK (base_url =
https://api.everyinfra.com/api/v1, same API key). It is free for accounts that have topped up;
per-minute limits follow cumulative top-ups (50 under $100, 500 from $100, 5,000 from $500, all keys
combined). It is meant for organizing the data you collect with EveryInfra.
Live numbers and activity
The badges read the live catalogs on every page view. The two charts are regenerated every day, from the same catalogs and from the git history of all EveryInfra repositories, including the private gateway (counts only); each chart prints its update time.
How a call flows
flowchart LR
A["Your agent"] --> B["MCP or REST"] --> C{"Key and balance"}
C -- "ok" --> D["Platform adapter"] --> E["Fixed JSON fields"]
C -- "no key" --> F["401, nothing charged"]
D -. "failed or empty" .-> G["Refund by charge_id"]
Status: what exists today
Layer
Repository state
What is not implied
Remote MCP
Production advertises 25 tools at https://api.everyinfra.com/mcp: capability list, data call and job status, AI chat, CAPTCHA types and solve, web search, plus account, usage, billing, API key, settings, webhook and ticket tools.
Discovery is not proof that an account is authenticated or has the balance or top-up a call requires.
Plugin package
Portable and host-specific manifests, skills, MCP metadata, and assets are present.
Presence is not a clean-host installation test.
Repository validator
Cross-host JSON, path, skill, MCP, and credential-safety checks are available.
Repository validation is not marketplace approval.
Self-hosted catalogs
Six marketplace manifests ( Codex, Claude Code, Cursor, GitHub Copilot, Qoder, CodeBuddy) and a Kimi Code root manifest are present in this repository.
A catalog file is not an official listing.
Market upload packages
MiniMax (CN and US), Qoder/ QoderWork skills and expert kit, and ClawHub skills are built from host-packages/ and pass scripts/build_host_packages.py check.
Not submitted to or approved by any of these markets.
OpenAI registered app
No .app.json or host-issued app ID is included.
Direct MCP metadata is not a registered ChatGPT connector.
Directories
Listed on Glama, LobeHub, Cursor Directory, ModelScope and the official MCP Registry (links under Community).
A directory listing is not an official Codex, Claude or Cursor marketplace approval.
WorkBuddy connectors
Connector packages for the WorkBuddy open platform are prepared under workbuddy/ and pass scripts/build_workbuddy.py check.
Not submitted to or approved by WorkBuddy.
Release claims should distinguish prepared, validated, installed, submitted, and
listed. Do not collapse them into one status.
Plugin install by host
An EveryInfra API key is required for paid operations; listing tools works without one. Keep the
key in the installing host's secret or environment configuration, never in a manifest, prompt,
committed file, or URL query string.
The explicit MCP entry is separate because a distributable plugin must not contain a real bearer
token. Start a new task after installing so Codex discovers the new skills and tools. A local clone
also works: codex plugin marketplace add <path-to-this-repository>. These commands are an
installation path, not evidence of an official Codex marketplace listing.
Claude Code
bash
claude plugin marketplace add everyinfra/everyinfra-agent-plugins
claude plugin install everyinfra@everyinfra
Configure the sensitive api_token value through Claude Code's plugin configuration flow. Avoid
putting a real token on a shell command line because shell history can retain it.
Cursor
For local development, copy or symlink plugins/everyinfra to
~/.cursor/plugins/local/everyinfra, reload Cursor, and set EVERYINFRA_API_KEY through Cursor's
plugin configuration. A team can import the repository through its own marketplace process. The
presence of .cursor-plugin/marketplace.json does not claim an official Cursor listing. Users who
only need tools can use the Install in Cursor button above.
GitHub Copilot and Agent Plugins 1.0 clients
The GitHub catalog is at .github/plugin/marketplace.json; the portable package starts at
plugins/everyinfra/plugin.json. Use the marketplace or local-plugin flow supported by your client.
Host support and secret injection differ, so confirm that the client reads EVERYINFRA_API_KEY
before making a paid call. The portable mcp.json deliberately contains no authorization header.
The catalog is .qoder-plugin/marketplace.json; the package manifest is
plugins/everyinfra/.qoder-plugin/plugin.json. Qoder loads the skills and the bundled .mcp.json,
which reads the key from EVERYINFRA_API_KEY. A local clone also works:
qoder plugins install ./plugins/everyinfra.
The catalog is .codebuddy-plugin/marketplace.json; the package manifest is
plugins/everyinfra/.codebuddy-plugin/plugin.json. CodeBuddy asks for the sensitive api_token
value when the plugin is enabled and sends it as the Authorization: Bearer header.
The root manifest .kimi-plugin/plugin.json loads the skills in plugins/everyinfra/skills/
and declares the remote MCP server at https://api-cn.everyinfra.com/mcp (same server and keys as
api.everyinfra.com, reachable from mainland China). The manifest carries no key, so listing tools
works out of the box; for paid calls, add the same server with an Authorization: Bearer header in
Kimi Code's MCP configuration. A bare repository URL installs the latest GitHub Release (0.4.0 and
later include this manifest); the /tree/main form installs the current main branch.
Markets that take uploads
MiniMax, Qoder/QoderWork and ClawHub want their own packages. Their manifests and listing metadata
live in host-packages/; the skill bodies always come from plugins/everyinfra/skills/.
python3 scripts/build_host_packages.py build writes the packages to dist/host-packages/ and does
not upload or submit anything. See host-packages/README.md for what each market receives.
Command line and SDKs
The Python package everyinfra on PyPI includes the everyinfra command-line tool and the
everyinfra-cli skill. The JavaScript / TypeScript client is the npm package everyinfra.
everyinfra skills install copies the everyinfra-cli skill into Claude Code or Codex when this
plugin is not installed. The sdk-v0.3.0 GitHub Release carries the same files as PyPI and npm,
with SHA256SUMS.
Repository layout
text
.agents/plugins/marketplace.json Codex self-hosted marketplace
.claude-plugin/marketplace.json Claude Code self-hosted marketplace
.cursor-plugin/marketplace.json Cursor self-hosted marketplace
.github/plugin/marketplace.json GitHub Copilot-compatible catalog
.qoder-plugin/marketplace.json Qoder CLI self-hosted marketplace
.codebuddy-plugin/marketplace.json CodeBuddy Code self-hosted marketplace
.kimi-plugin/plugin.json Kimi Code root manifest (install from the repository URL)
plugins/everyinfra/plugin.json Agent Plugins 1.0 package manifest
plugins/everyinfra/.<host>-plugin/plugin.json Codex, Claude Code, Cursor, Qoder and CodeBuddy manifests
plugins/everyinfra/mcp*.json Portable and host-specific MCP metadata
plugins/everyinfra/skills/ Router, product, and outcome skills
plugins/everyinfra/assets/ Plugin icon and light/dark logos
assets/ README hero images and platform icons
i18n/ README in Simplified Chinese and Spanish
workbuddy/ WorkBuddy connector packages (MCP + Skill)
host-packages/ MiniMax, Qoder/QoderWork and ClawHub package sources
llms-install.md Install steps for agents that set up MCP servers
Run host validators that are available in your environment and report unavailable validators as
skipped, not passed, for example claude plugin validate plugins/everyinfra --strict.