MCP servers for data & analytics

by Apify

Official Apify MCP server exposing thousands of ready-made Actors (scrapers/crawlers/automations) from the Apify Store as tools: agents can discover Actors, run them, and access storages and results; also gives coding assistants access to Apify docs.

Capabilities: discover and run Apify Actors · scrape social media, search engines, maps, e-commerce sites · access datasets/storages and run results · Apify documentation access

Authentication

Apify API token or OAuth on the hosted server

Use cases

e-commerce price monitoring, review scraping, lead research

⚠ Dynamic tool discovery means an agent can run arbitrary paid Actors — set cost controls; Actor quality varies by author.

Documentation / source →

by Google Cloud

Two official Google routes to BigQuery over MCP: a fully managed remote BigQuery MCP server, and the open-source self-hosted MCP Toolbox for Databases (supports BigQuery, AlloyDB, Cloud SQL, Postgres and more via tools.yaml configuration).

Capabilities: list datasets and tables · get table schemas · execute SQL queries · multi-database tool configuration (Toolbox)

Authentication

Google Cloud IAM (OAuth / Application Default Credentials)

Use cases

data agents on the warehouse, Claude Code or Gemini CLI directly on BigQuery datasets, agent development with ADK

⚠ Standard BigQuery query costs apply; restrict access via IAM and prefer read-only datasets.

Documentation / source →

by Brave Software

Official Brave MCP server for the Brave Search API: web, local/POI, image, video and news search plus LLM context/summarization, over STDIO and HTTP transports.

Capabilities: web search · local points-of-interest search · image search · video search · news search

Authentication

Brave Search API key (free tier available)

Use cases

privacy-friendly current web results in agents, news monitoring

⚠ The old reference implementation in modelcontextprotocol/servers is archived; use the vendor repo.

Documentation / source →

by ClickHouse

Official MCP server for ClickHouse (self-hosted or ClickHouse Cloud): run SQL queries, list databases and tables with pagination, plus an embedded chDB engine to query files, URLs and databases directly without ETL.

Capabilities: execute SQL queries on ClickHouse clusters · list databases and tables · query files/URLs directly via embedded chDB engine

Authentication

ClickHouse credentials (host/user/password); read-only by default (CLICKHOUSE_ALLOW_WRITE_ACCESS=false)

Use cases

real-time analytics and event data (clickstream, e-commerce events) via agents, log analysis

⚠ Enable write access only deliberately; queries run against production compute.

Documentation / source →

by Databricks

Official ready-to-use hosted MCP servers connecting agents to Unity Catalog data and functions, AI/Vector Search indexes, Genie Spaces, and SaaS connectors (Slack, GitHub), centrally managed via Unity AI Gateway.

Capabilities: Unity Catalog data and functions as tools · Vector Search indexes · Genie Spaces (genie_ask / genie_poll_response) · SaaS app connectors

Authentication

Databricks OAuth with on-behalf-of-user auth; Unity Catalog permissions enforced

Use cases

enterprise data agents with existing governance, lakehouse Q&A via Genie, custom UC functions as agent tools

⚠ Requires a Databricks workspace (AWS/Azure/GCP); custom MCP servers can additionally be hosted on Databricks Apps.

Documentation / source →

by dbt Labs

Official dbt MCP server with 60+ tools across 9 toolsets: dbt command execution, project metadata/lineage discovery, Semantic Layer querying (metrics, dimensions, entities, compiled SQL), SQL execution, job management, and docs search. Local (uvx) and remote hosted flavors.

Capabilities: run dbt commands (Core/Fusion/Cloud CLI) · discovery: project metadata and lineage · semantic layer: get_metrics/get_dimensions/get_entities/query_metrics · execute_sql on dbt platform · job management

Authentication

local with dbt credentials; remote with dbt platform service tokens

Use cases

reliable metric answers via governed semantic layer, dbt development with agents, lineage questions

⚠ Enable toolsets selectively (60+ tools is heavy context); semantic layer tools require a dbt platform account.

Documentation / source →

by Elastic

Official MCP endpoint of Elastic Agent Builder (Elasticsearch 9.2+ and Serverless): exposes built-in and custom tools that query Elasticsearch indices to external clients like Claude Desktop, Cursor and LangChain; supersedes the deprecated standalone Elasticsearch MCP server.

Capabilities: search and ES|QL queries over indices · built-in and custom Agent Builder tools · expose agents via MCP, A2A and REST

Authentication

Elasticsearch API key with Kibana Agent Builder privileges

Use cases

search/RAG over own indices, querying observability and security data via agents

⚠ Requires Elasticsearch 9.2+ or Serverless. Legacy github.com/elastic/mcp-server-elasticsearch is DEPRECATED (critical security updates only).

Documentation / source →

by Exa Labs

Official Exa MCP server (remote at https://mcp.exa.ai/mcp) providing neural/semantic web search with domain, date and category filters, web crawling, code search and deep research; a separate Websets MCP server also exists.

Capabilities: semantic web search with filters · web crawling · code search · deep research

Authentication

Exa API key

Use cases

research agents, semantic queries where keyword search fails, structured company/people research

⚠ Paid per API usage.

Documentation / source →

by Firecrawl

Official Firecrawl MCP server (npx -y firecrawl-mcp) adding web scraping and search: single-URL scrape, batch scrape, site mapping, search, async crawling, structured extraction and status checks.

Capabilities: scrape single URL · batch scrape · map website structure · web search · async crawl

Authentication

Firecrawl API key

Use cases

e-commerce competitor and price monitoring, content to markdown for RAG, site migrations

⚠ Cloud API with credits; respect robots.txt and target-site terms.

Documentation / source →

by MotherDuck

Official MotherDuck MCP server for analytical SQL with DuckDB: local DuckDB files, in-memory databases, S3-hosted databases and MotherDuck cloud; read/write queries and catalog browsing. A remote MCP server launched in December 2025.

Capabilities: execute SQL read/write queries via DuckDB · browse database catalogs · query local files, S3 and MotherDuck cloud

Authentication

local: none (file access); MotherDuck: service token; read-only modes available

Use cases

ad-hoc analysis of CSV/Parquet/S3 data, lightweight analytics without a warehouse

⚠ Read-write mode can mutate data — use read-only flag for exploration.

Documentation / source →

by Snowflake

Official Snowflake-managed MCP server (GA since 2025-11-04) serving Cortex Analyst, Cortex Search, Cortex Agents, custom tools and SQL execution without separate infrastructure; an official open-source variant (Snowflake-Labs/mcp) adds object management and SQL orchestration.

Capabilities: Cortex Analyst (natural language to SQL on governed data) · Cortex Search (semantic search over documents) · Cortex Agents · SQL execution · custom tools

Authentication

Snowflake built-in OAuth with RBAC at server and tool level; PATs supported

Use cases

data agents querying governed warehouse data, natural-language BI, RAG over Snowflake documents

⚠ Costs run through Snowflake compute; configure RBAC before broad rollout.

Documentation / source →

by John Blazick (blazickjp, community)

Community MCP server bridging AI assistants and the arXiv research repository: search papers by keyword/author/category, retrieve metadata, download and analyze paper content.

Capabilities: search arXiv papers · retrieve paper metadata · download papers · analyze/summarize paper content

Authentication

none (public arXiv API)

Use cases

literature research, tech trend analysis, research assistants

⚠ Community project; respect arXiv rate limits. Also distributed as Docker image mcp/arxiv-mcp-server; alternative implementations exist.

Documentation / source →

by Rudra-ravi (community)

Community MCP server that searches Wikipedia and retrieves article content, summaries, sections, links and related topics for factual grounding; multi-language support (including Dutch) via a language argument. No official Wikimedia server exists.

Capabilities: search Wikipedia · retrieve full articles as Markdown · multilingual lookups

Authentication

none (public Wikipedia API)

Use cases

fact checking, background knowledge in research agents

⚠ No official Wikimedia server; multiple community implementations with varying feature sets — vet the specific implementation.

Documentation / source →