5 tools. Every endpoint.

Name: Anysite MCP Server
Rating: 4.9 (89 reviews)
Author: Anysite

The Anysite MCP Server gives your AI assistant structured access to hundreds of endpoints across LinkedIn, Instagram, Twitter, Reddit, YouTube, and more — through 5 universal tools.

5 universal tools access every endpoint on the platform — current, future, and custom. No hardcoded tool names, no manual updates.

Get MCP Access Read the docs

Use code MCP30 for your first month free.

discover // find endpoints, read schemas

execute // fetch data from any endpoint

get_page // paginate cached results

query_cache // filter, aggregate, sort server-side

export_data // download as JSON, CSV, JSONL

meta-tools

100s

endpoints

40+

data sources

~1,000

tokens at startup

$30/mo

MCP Unlimited

How the 5 tools work

Every request flows through the same sequence. Start with discover to find what's available. Call execute to fetch data. Use get_page, query_cache, and export_data to work with what comes back.

discover

find endpoints

→

execute

fetch data

→

get_page

paginate

→

query_cache

filter & aggregate

→

export_data

download files

discover

Find what endpoints exist and what parameters they take

Returns the full endpoint catalog with parameter schemas for any platform. You don't need to know what's available — the tool tells you.

discover("linkedin", "user")
→ endpoint: linkedin/user
  params: user (string, required), include_skills (bool)
  description: Full profile data including experience, education, skills

execute

Fetch data from any endpoint

Pass a platform, endpoint, and parameters. Results land in a ClickHouse cache. First page returns immediately; paginate the rest with get_page.

execute("linkedin", "user", {"user": "darioamodei"})
→ {name: "Dario Amodei", headline: "...", experience: [...]}
  cache_key: "lk_usr_a7f3bc"
  total_results: 1

get_page

Page through large result sets without new API calls

Move through large result sets from cache. No additional requests to the source platform.

get_page("lk_search_9d42ef", offset=10, limit=10)
→ profiles 11-20 from cache

query_cache

Filter, aggregate, and sort results server-side

Run SQL-style queries against cached data. Filter, aggregate, sort — get a targeted dataset without loading everything into context.

query_cache("lk_search_9d42ef",
  conditions=[{"field": "location", "op": "contains", "value": "San Francisco"}],
  sort_by="followers", sort_order="desc", limit=25)
→ 25 SF-based profiles sorted by follower count

export_data

Download full datasets as JSON, CSV, or JSONL

Export any cached result set to a file. Download URL valid for 7 days.

export_data("lk_search_9d42ef", "csv")
→ download_url: "https://api.anysite.io/export/..."
  format: csv
  rows: 200
  expires: 7 days

Why 5 tools are enough

Most MCP servers ship one tool per endpoint. That doesn't scale — each definition eats context tokens, the LLM picks wrong tools from long lists, and new endpoints require new code.

Anysite loads 5 tool definitions at startup (~1,000 tokens). The endpoint catalog lives on the server via OpenAPI. New data sources appear automatically. Custom endpoints appear automatically. The tools never change.

Capability	How it works
Endpoint access	Hundreds of endpoints across 40+ platforms — all through `discover` + `execute`
Context overhead	~1,000 tokens at startup. Your AI keeps its context for actual work
New data sources	Appear automatically via OpenAPI spec. No server changes needed
Custom endpoints	Build your own through the Custom Endpoint Builder. Available in MCP immediately
Pagination	Built in via `get_page`. Page through cached results without new API calls
Server-side analysis	Filter, aggregate, sort cached data via `query_cache`
Data export	Download full datasets as JSON, CSV, or JSONL via `export_data`

5 tools. Hundreds of endpoints. Zero maintenance.

Server-side analysis

Every execute() call caches results in ClickHouse for 7 days. query_cache runs filters, aggregations, and sorts against that cache server-side — your AI gets targeted data, not a raw dump of 200 records.

Filter by location and followers

query_cache("cache_key", conditions=[
  {"field": "location", "op": "contains", "value": "San Francisco"},
  {"field": "followers", "op": ">", "value": 500}
])

Returns only profiles in San Francisco with 500+ followers. The rest stay in cache.

Aggregate: average followers by industry

query_cache("cache_key",
  aggregate={"field": "followers", "op": "avg"},
  group_by="industry"
)

Returns a table: industry to average follower count. No raw data in context.

Top 10 by metric

query_cache("cache_key",
  sort_by="followers",
  sort_order="desc",
  limit=10
)

Returns the 10 highest-follower profiles from a search of 200.

Export the filtered set

export_data("cache_key", "csv")

Downloads the full filtered result as a CSV file. Valid for 7 days.

Filter operators: = != > < >= <= contains not_contains

Aggregation functions: sum avg min max count uniq

Hundreds of endpoints across 40+ sources

Professional networks, social platforms, business directories, search engines, and the open web. All through the same 5 tools.

Professional

LinkedIn 31 endpoints — profiles, companies, jobs, posts, search, email, messaging

Social

Twitter/X — user profiles, tweets, search, followers
Instagram — profiles, posts, reels, search
Reddit — posts, comments, subreddits, search
YouTube — videos, channels, search, subtitles
Pinterest

Business Intelligence

Crunchbase — company profiles
SEC EDGAR — filings, company search
Glassdoor — company reviews
G2 — software reviews
TrustPilot — business reviews
BuiltWith — technology stack data
Capterra — software reviews

Startup Ecosystem

Y Combinator — company search, founders, details
AngelList — startup profiles
ProductHunt — product launches

Search

DuckDuckGo — web search
Google News — news search
Google Maps — location data
HackerNews — posts and comments
GitHub — repos and user data

Web

Web Parser — any URL to structured data
Sitemap Parser — full site crawls

Other

Amazon — product data
TripAdvisor — reviews, locations

Custom endpoints built through the Custom Endpoint Builder appear in MCP automatically. No server updates needed. You add an endpoint to the platform, and it's available to every connected AI assistant within minutes.

Connect in 60 seconds

Three setup paths depending on your client. Claude Desktop uses OAuth — no API keys. Claude Code and Cursor use an API key in a single config line.

Go to anysite.io and open the MCP Server integration page
Copy the OAuth URL: https://mcp.anysite.io/mcp
In Claude Desktop: Settings → Connectors → Add Custom Connector
Paste the URL and click Connect
Authorize in your browser (one-time)

One required setting: In Claude Desktop: Settings → Capabilities → Tool access → select "Tools already loaded".

This ensures the 5 meta-tools are always in context. The default "Load tools when needed" mode uses name-based matching that doesn't work well with generic names like discover and execute.

Verify: Ask Claude: "What MCP tools do you have?"

claude mcp add --transport sse anysite \
  "https://api.anysite.io/mcp/direct?api_key=YOUR_KEY"

Verify:

claude mcp list
# anysite     connected

Add to ~/.cursor/config.json:

{
  "mcpServers": {
    "anysite": {
      "url": "https://api.anysite.io/mcp/direct?api_key=YOUR_KEY",
      "transport": "sse"
    }
  }
}

Verify: Open Cursor, type /mcp to see available tools.

Get API Key

What you can do with it

These are working examples, not hypotheticals. Each shows the natural language request, the tools it triggers, and what you get back.

Lead Generation

"Find CTOs at Series A fintech startups in New York with AI or ML background."

1. discover — finds the search/users endpoint and its filters
2. execute — runs the search, caches 100 results
3. query_cache — filters server-side by seniority = C-Level
4. export_data — downloads the filtered list as CSV

Structured list of CTO profiles with name, company, current role, LinkedIn URL, and email where available.

Competitive Intelligence

"Get the last 50 posts from @competitor on LinkedIn and Twitter. Summarize their product announcements and engagement metrics."

1. execute — fetches LinkedIn company posts
2. execute — fetches Twitter posts
3. query_cache — aggregates avg engagement by post type
4. LLM summarizes with engagement data already loaded

Summary of announcements by topic, average engagement by post type, notable spikes flagged.

Recruiting

"Search LinkedIn for senior Python developers with Kubernetes experience in Berlin. Show current companies and how long they've been in their roles."

1. execute — searches with keywords and location, returns up to 200 profiles
2. query_cache — filters to >12 months in current role, sorts by tenure

Ranked list of candidates with current role, company, tenure, and LinkedIn URL.

Market Research

"Search Reddit and HackerNews for discussions about our product category from the last 30 days. Identify the top pain points."

1. execute — searches Reddit with time filter
2. execute — searches HackerNews (last 30 days)
3. query_cache — filters by comment count, sorts by recency
4. LLM extracts pain points from the filtered dataset

Summary of recurring pain points by frequency, with source posts linked.

Pricing

MCP Unlimited is the straightforward option: $30/month for unlimited MCP requests with no credit counting. Credit-based plans include MCP access if you also need the REST API or CLI.

Recommended for MCP

MCP Unlimited

$30/month

All 5 meta-tools
Hundreds of endpoints across 40+ platforms
Unlimited requests (fair use: 50K/month)
Server-side analysis (query_cache)
Data export (JSON/CSV/JSONL)
Person Analyzer skill
Competitor Analyzer skill
OAuth or API key auth
Works with Claude Desktop, Claude Code, Cursor

Rate limit: 1 request per 10 seconds

Start with MCP Unlimited

MCP Unlimited covers MCP access only. If you also need the REST API or CLI, credit-based plans include MCP access.

Need REST API or CLI too?

Plan	Price	Credits	Rate Limit
Starter	$49/mo	15K credits	60 req/min
Growth	$200/mo	100K credits	90 req/min
Scale	$300/mo	190K credits	150 req/min
Pro	$549/mo	425K credits	200 req/min
Enterprise	$1,199+/mo	1.2M+ credits	200 req/min

All credit-based plans include full MCP access.

Starter plan includes a 7-day free trial with 1,000 credits.

PAYG top-ups available at $2.90 per 1K credits (minimum $20, requires active subscription).

First month free with code MCP30. Apply at checkout.

MCP to explore. CLI to execute.

MCP and the CLI run on the same engine. MCP is conversational — describe what you want, get targeted results. CLI is operational — YAML pipelines, batch jobs, scheduled collection at scale.

	MCP Server	Anysite CLI
Interface	Natural language via AI assistant	YAML pipelines, terminal
Best for	Exploration, ad-hoc queries, research	Production pipelines, batch processing
Setup	Connect in 60 seconds	pip install, one config file
Data volume	Single queries, filtered results	Thousands of records, parallel execution
Storage	ClickHouse cache, 7-day export	Any database (SQLite, PostgreSQL, ClickHouse)
Scheduling	Not built in	Cron scheduling included
LLM enrichment	Via your AI assistant	Built-in (classify, summarize, enrich)
Pricing	$30/mo unlimited	Credit-based plans from $49/mo

Most teams use both: MCP to understand what data is available and prototype a workflow, CLI to run that workflow in production on a schedule.

Read about the Anysite CLI

Common questions

What is the Anysite MCP Server?

A cloud-hosted Model Context Protocol service that gives AI assistants like Claude, Cursor, and ChatGPT structured access to web data. You describe what you need in natural language; the server fetches structured JSON from LinkedIn, Twitter, Reddit, and 40+ other sources.

How does the MCP Server architecture work?

Five meta-tools — discover, execute, get_page, query_cache, export_data — access all endpoints dynamically. The endpoint catalog loads from the OpenAPI spec at startup, consuming roughly 1,000 tokens. New endpoints appear automatically without server changes. Server-side filtering and aggregation via query_cache, pagination via get_page, and data export via export_data keep your AI's context clean.

How do I set up the MCP Server with Claude Desktop?

Go to anysite.io, copy the OAuth URL (https://mcp.anysite.io/mcp), and add it as a custom connector in Claude Desktop under Settings → Connectors. One important step: in Settings → Capabilities → Tool access, select "Tools already loaded" — the default setting doesn't work well with generic tool names like discover and execute.

What does MCP Unlimited include?

All 5 meta-tools, hundreds of endpoints, unlimited requests (fair use 50K/month), server-side analysis, data export, and both pre-built skills (Person Analyzer, Competitor Analyzer). Covers MCP access only — REST API and CLI requests require a credit-based plan.

Can I use this with Cursor or other MCP clients?

Yes. The server uses standard MCP 1.0 over SSE transport. Any MCP-compatible client works. Cursor setup is a single JSON config file. See the setup section for exact configuration.

What is query_cache and why does it matter?

query_cache runs filter, aggregation, and sort queries against your cached results server-side — before any data loads into your AI's context. Instead of loading 200 raw profiles to find 15 matching ones, you send a query to the cache and receive only the 15. It keeps context windows clean and makes large dataset workflows practical.

Do custom endpoints I build appear in MCP automatically?

Yes. Endpoints built with the Custom Endpoint Builder are registered in the OpenAPI spec. The MCP server loads the spec dynamically, so custom endpoints appear in discover() and are callable via execute() without any changes to the MCP server itself.

Is there a free plan?

No. The free tier was discontinued on March 1, 2026. Starter plan ($49/mo) includes a 7-day free trial with 1,000 credits. MCP Unlimited ($30/mo) has no trial period.

What platforms does the MCP Server support?

LinkedIn (31 endpoints), Instagram, Twitter/X, Reddit, YouTube, Pinterest, HackerNews, DuckDuckGo, Google News, Google Maps, GitHub, SEC EDGAR, Y Combinator, Crunchbase, Glassdoor, G2, TrustPilot, Capterra, BuiltWith, Amazon, TripAdvisor, AngelList, ProductHunt, web parser (any URL), sitemap parser, and custom endpoints.

How does authentication work?

Claude Desktop supports OAuth — no API keys to manage. Claude Code and Cursor use an API key in a single config line. Both methods are covered in the setup section. OAuth is recommended when available because it handles token rotation automatically.

Start querying the web

MCP Unlimited gives your AI assistant access to hundreds of endpoints across LinkedIn, Instagram, Twitter, and 40+ other sources. Five tools. No endpoint management. No per-request counting.

First month free with code MCP30.

Get MCP Access Read the docs