EveryInfra MCP Server & Agent Plugin
One API key gives AI agents structured public data from 106 platforms, web search and CAPTCHA solving.
- Source repo
- everyinfra/everyinfra-agent-plugins
- Stars
- ★ 649
- Last updated
- 1d ago
- License
- Apache-2.0
- Primary language
- Python
- FA score
- 55/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude Code · OpenAI APIChatGPT (Needs adaptation)
- Cost
- Needs a paid subscription or licence
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A product marketer uses Claude Code to pull the top 30 Amazon results for "standing desk" and compare price, rating and review count.
- Not a fit if
- Teams that must run fully offline and cannot send requests to a vendor cloud
- Developers who want a completely free open-source scraper with no per-call billing
- Projects that only need plain static HTML scraping without platform-native structured fields
- Website
- everyinfra.com/en/mcp-server ↗ set on GitHub
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-20latest release sdk-v0.3.0 (2026-11-01)+24.6 ★/day over 7 days
- Source review
- 55/100 · Major gaps
What does this agent do, and when should you use it?
EveryInfra is a remote MCP server plus an installable agent plugin that puts public web data collection, web search and CAPTCHA solving behind a single API key. The repository is not the backend: it ships 11 discoverable skills (one router, seven product skills, two outcome workflows and an everyinfra-cli skill), MCP connection metadata pointing at https://api.everyinfra.com/mcp, and host-specific manifests for Codex, Claude Code, Cursor, Qoder, CodeBuddy and Kimi Code. At runtime the hosted server advertises 25 tools covering capability listing, data calls and job status, AI chat, CAPTCHA types and solving, web search, plus account, usage, billing, API key and ticket tools. Calls are billed per use with failed or empty results not charged, starting from $0.56 per 1,000 calls. Beyond MCP, the same key works over REST, the everyinfra CLI on PyPI and the everyinfra npm client, so scripts and CI jobs can use it too.
A user asks for an outcome in plain words; the host agent picks the router skill or a product skill and calls MCP tools such as everyinfra_call_api or everyinfra_chat. Discovery comes first: the agent lists live capabilities and tool schemas before inventing parameters or estimating cost, then executes the platform call and receives fixed JSON fields. The skills split into three layers — everyinfra routes when the user names the product, needs several products or has not chosen one; everydata, everysearch, everyai, everysolve, everynumber, everymail and everyproxy cover data, search, chat, CAPTCHA, numbers, email and proxies; everyinfra-crosscheck-research and everyinfra-bulk-data-export produce evidence-backed answers and bounded, deduplicated CSV or JSON exports. The CLI path (pipx install everyinfra) adds dry-run cost previews, a budget cap and file output. Each call flows through key and balance check, platform adapter and fixed JSON fields; a missing key returns 401 with nothing charged, and failed or empty results are refunded by charge_id.
- A product marketer uses Claude Code to pull the top 30 Amazon results for "standing desk" and compare price, rating and review count.
- A fact-checking researcher asks the agent to find current sources for a claim through the web search tools and cross-check them into a cited answer.
- An analyst bulk-collects public Xiaohongshu, TikTok or YouTube posts and comments and runs the bulk export workflow to get a bounded, deduplicated CSV.
- A developer writing a scraper in Cursor calls the MCP tools to confirm response fields instead of rewriting per-platform parsing code.
- An automation engineer runs
everysolveagainst a target they are authorized to automate, after listing CAPTCHA types to confirm coverage. - A CI job or shell script uses the
everyinfraCLI with a budget cap to run scheduled pulls, previewing cost with a dry run first. - A team that already collected comments uses
everyinfra_chatthrough an OpenAI-compatible endpoint to clean and label them into JSON. - A team inside mainland China points Kimi Code at
api-cn.everyinfra.com, which serves the same backend with the same keys.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
pipx install everyinfra # command-line tool (or: uv tool install everyinfra)npm install everyinfra
- Environment variables
EVERYINFRA_API_KEY- Config files
~/.cursor/plugins/local/everyinfra
How do you install or deploy this agent?
Get an API key at everyinfra.com first; listing tools works without one. Claude Code can add the remote MCP server directly:
claude mcp add --transport http everyinfra https://api.everyinfra.com/mcp \
--header "Authorization: Bearer <API Key>"Codex injects the key from an environment variable:
export EVERYINFRA_API_KEY="<API Key>"
codex mcp add everyinfra --url https://api.everyinfra.com/mcp \
--bearer-token-env-var EVERYINFRA_API_KEYFor the full plugin (skills plus manifests) with Codex:
codex plugin marketplace add everyinfra/everyinfra-agent-plugins
codex plugin add everyinfra@everyinfra
codex mcp add everyinfra \
--url https://api.everyinfra.com/mcp \
--bearer-token-env-var EVERYINFRA_API_KEYClaude Code plugin install:
claude plugin marketplace add everyinfra/everyinfra-agent-plugins
claude plugin install everyinfra@everyinfraQoder CLI:
qoder plugins marketplace add everyinfra/everyinfra-agent-plugins
qoder plugins install everyinfraCodeBuddy Code runs /plugin marketplace add everyinfra/everyinfra-agent-plugins and /plugin install everyinfra@everyinfra inside the session; Kimi Code uses /plugins install https://github.com/everyinfra/everyinfra-agent-plugins/tree/main. Cursor and VS Code can use the install buttons at the top of the README, or copy/symlink plugins/everyinfra to ~/.cursor/plugins/local/everyinfra and set EVERYINFRA_API_KEY in the plugin configuration.
Command line and SDKs:
pipx install everyinfra # command-line tool (or: uv tool install everyinfra)
everyinfra doctor # checks version, network, key and config
python -m pip install everyinfra
npm install everyinfraKeep the key in the host's environment or secret configuration, never in a manifest, prompt, committed file or URL query string. Start a new Codex task after installing so it discovers the new skills.
How do you use this agent?
Once installed with a key configured, describe the outcome in plain words and let the agent pick the router or product skill. For example, in Claude Code or Codex:
Find the top 30 Amazon results for "standing desk" and compare price, rating and review count.The agent lists capabilities before calling everyinfra_call_api. AI text generation maps to the everyinfra_chat MCP tool, or to REST through any OpenAI SDK:
from openai import OpenAI
client = OpenAI(
base_url="https://api.everyinfra.com/api/v1",
api_key="<API Key>",
)
resp = client.chat.completions.create(
model="<model name from the live catalog>",
messages=[{"role": "user", "content": "Label these comments as positive or negative."}],
)
print(resp.choices[0].message.content)Purchases, sends, rentals, CAPTCHA solves and credential delivery are 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, and read docs/permissions.md and docs/rest-only.md before paid or state-changing calls.
What are this agent's strengths and limitations?
- One key and one MCP endpoint cover 106 platforms, 573 data APIs, 17 search tools and 52 CAPTCHA types, so agents need no per-platform auth setup.
- Responses across platforms share a fixed JSON shape, and failed or empty calls are refunded by charge_id, which limits the cost of dead runs.
- Four documented access paths — MCP, plugin, CLI and REST/SDK — with plugin manifests for Codex, Claude Code, Cursor, Qoder, CodeBuddy and Kimi Code.
- The repo ships cross-host validators (
scripts/validate.py,build_host_packages.py check,build_workbuddy.py check,build_release.py --check-source) so you can verify manifests and credential safety before distributing. - The CLI offers dry-run cost previews and a budget cap, which makes it viable inside scheduled automation.
- The backend is a hosted commercial API: you need network access and a paid EveryInfra key, and the gateway source is not in this repo, so the capability cannot be self-hosted.
- Pricing starts at $0.56 per 1,000 calls but rates and availability are governed by the live catalog; the README numbers are timestamped and may drift.
- EveryNumber, EveryMail and EveryProxy are documented as REST guidance only, and CAPTCHA solving carries authorization risk that adopters must assess themselves.
- Several distribution channels are marked prepared or validated rather than submitted or listed — market upload packages, WorkBuddy connectors and the Kimi manifest are not marketplace approvals.
- With Codex, a distributable plugin cannot carry a real bearer token, so an extra explicit
codex mcp addstep is required beyond the plugin install.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| EveryInfra MCP Server & Agent Plugin This agent | 55 · Major gaps | MCP serverPaid | ★ 649 | 1d ago | Python | Codex · Claude Code · OpenAI API |
| Bright Data MCP | 62 · Some gaps | MCP serverFreemium | ★ 2.7k | 3d ago | JavaScript | Claude Code · Claude.ai |
| Maxun | 44 · Major gaps | Hosted serviceFreemium | ★ 19k | 1d ago | TypeScript | — |
| webclaw Web Extraction Engine | 80 · Good | CLIFreemium | ★ 2.4k | 2d ago | Rust | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Permission and side-effect boundaries are well documented: docs/permissions.md separates read/discovery from paid and external actions and requires user authorization for captcha, number, email, and proxy operations; keys are injected via env vars or host secrets and forbidden in manifests/URLs/prompts. Deductions: dependency security relies only on CI validation scripts with no dependency manifest or vulnerability scanning; rollback only mentions uninstalling the plugin, with no refund/recovery detail for failed paid calls; source attribution points only to everyinfra.com without data-origin or platform-compliance notes.
The CLI skill provides a full exit-code table and error JSON shape, so failure messaging is solid. However README numbers are internally inconsistent (573 APIs/106 platforms vs later 'All 90 platforms'), and most capability depends on a remote MCP endpoint and an external pipx package whose availability cannot be confirmed statically; no version pinning or offline degradation is described.
Targets many hosts (Codex/Claude/Cursor/Copilot/Qoder etc.) with install paths, a clear capability matrix, and explicit notes that EveryNumber/EveryMail/EveryProxy remain REST guidance rather than MCP. Deductions: trigger precision depends on skill descriptions with no conflict-resolution rules; environment fit does not cover offline or restricted networks.
Information architecture is layered (README/docs/skills/manifests), install notes are detailed, known limitations (not submitted to marketplaces, not official listings) are honestly stated, and the Apache-2.0 license is complete. Deductions: version 0.4.0 appears only in manifests with no CHANGELOG; naming is inconsistent between EveryInfra/EveryData and io.github.xiaoxihexiaoyu; maintenance responsibility points to an unverified publisher with no SLA or contact.
Output usability is good: CLI supports JSON/JSONL/field selection/file output, and the bulk-export skill includes dedup and verification steps. Marginal value is limited because the core value is a paid third-party API wrapper and the repository contains no backend; cost control exists via dry-run and budget caps, but pricing and billing depend on a remote catalog that cannot be verified statically.
Docs repeatedly defer to the live catalog and MCP schema and distinguish prepared/validated/installed/submitted/listed, so fact/inference separation is decent. Deductions: key figures (573 APIs, 106 platforms, $0.56/1K) are self-asserted with no in-repo data to check; cross-source corroboration relies on external directory links with no verifiable samples inside the package.
- Publisher identity is unverified by FollowAgents; maintenance responsibility and update path are unclear, so verify provenance before installing.
- Core capability depends on a remote MCP endpoint and an external pipx package; static review cannot confirm availability, billing, or data compliance.
- README platform/API counts are inconsistent (106 vs 90) and key figures are self-asserted; do not cite them directly.
- Paid calls, captcha, number, email, and proxy actions require explicit user authorization to avoid irreversible side effects.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
sdk-v0.3.0 SDK v0.3.0
Release notes excerpt
EveryInfra SDK v0.3.0. The Python package everyinfra 0.3.0 is on PyPI (pip install everyinfra). It now includes the everyinfra command-line tool: find a capability in the catalog, read its contract, preview a call and its cost with --dry-run, cap a batch with --max-cost-usd, and run platform data, web search, CAPTCHA solving and the AI API with one key. everyinfra skills install adds the everyinfra-cli skill for coding agents. …View on GitHub ↗ -
v0.4.0 v0.4.0 — More agent hosts, tagged links, README first screen
Release notes excerpt
EveryInfra agent plugins v0.4.0 One remote MCP server and ten skills for structured data from TikTok, Instagram, Amazon, YouTube, X, Reddit, LinkedIn, Google SERP, Xiaohongshu (RedNote) and Douyin: 414 data capabilities across 90 platforms, 17 web search tools and 52 CAPTCHA types. Pay per use from $0.56 per 1,000 data calls; failed or empty calls are not charged. Highlights - Install from more agent hosts: Qoder CLI, CodeBuddy Code and Kimi Code read their manifests straight from this repository (Kimi: /plugins install https://github.com/everyinfra/everyinfra-agent-plugins). …View on GitHub ↗ -
v0.3.0 v0.3.0 — Cleanup route retired
Release notes excerpt
EveryInfra agent plugins v0.3.0 Highlights - Retire the data cleanup route: the two cleanup tools were removed from the remote MCP server on 2026-10-09. Cleaning, labeling, extraction and summaries of collected data now go to the AI API (everyinfra_chat or POST /api/v1/chat/completions), free for accounts that have topped up. - Remove the cleanup reference and cleanup examples from the README, skills, manifests and discovery files. - Update server.json for the official MCP Registry entry io.github.xiaoxihexiaoyu/everyinfra …View on GitHub ↗
FAQ
Can I use the plugin without the MCP server, or the MCP server without the plugin?
everyinfra in a skill list does not by itself prove the MCP connection is authenticated.Is there a free tier?
everyinfra_chat is free for accounts that have topped up, with per-minute limits tied to cumulative top-ups (50 under $100, 500 from $100, 5,000 from $500 across all keys).Will I be charged when a call fails?
Can the agent just buy numbers, send email or order proxies on its own?
Does it work from mainland China?
Is EveryInfra MCP Server & Agent Plugin open source?
Is EveryInfra MCP Server & Agent Plugin safe to use?
Is EveryInfra MCP Server & Agent Plugin still maintained?
Who makes EveryInfra MCP Server & Agent Plugin?
What is EveryInfra MCP Server & Agent Plugin's official website?
Thanks — we use this to improve the page.