Best Data Analytics Plugins, Skills, and MCP Servers in 2026
Research-backed data and analytics plugins, skills, and MCP servers with platform availability and direct setup routes.
Updated July 25, 2026
These extensions let agents reach real databases, warehouses, BI platforms, and product-analytics tools, then run queries and follow repeatable analysis workflows. The main decision is how much access to grant: broad exploration is convenient, but production data usually deserves scoped, read-only credentials.
The official MongoDB package brings together the MongoDB MCP Server and reusable skills for schemas, queries, aggregations, indexes, connections, and Atlas workflows, giving an agent both live database access and current MongoDB practice.
The result is an agent that inspects and works with MongoDB while grounding its code and database advice in MongoDB’s current guidance rather than generic patterns. Keep access read-only and least-privilege until a task clearly needs writes.
Google’s official, experimental local MCP server for the Google Analytics Admin and Data APIs. It exposes read-only tools to list accounts and properties, inspect custom dimensions and metrics, and run core, funnel, and realtime GA4 reports.
Use it to answer analytics questions - traffic, conversions, funnels, and realtime activity - directly against GA4 instead of exporting reports by hand. Access is read-only by scope, so it reports on data but cannot change GA configuration. Marketers who reach it from the Marketing & SEO guide land here, where it is ranked among data and analytics extensions.
A local MCP gateway for PostgreSQL, MySQL, MariaDB, SQL Server, and SQLite. It deliberately exposes just two core tools, schema and object search plus SQL execution, alongside optional parameterized custom tools.
One token-efficient interface covers several database engines, and the guardrails are the real draw: read-only mode, row limits, query timeouts, and TLS/SSH support. Use a dedicated least-privilege account regardless, and keep the server off untrusted networks.
An integration that pairs Neon’s Serverless Postgres skill with the Neon MCP Server for live project and database work. The agent can provision databases, create branches, run queries and migrations, and validate connections while following current Neon patterns instead of generic Postgres advice.
Reach for it when an agent needs to build against or manage a Neon database. Neon positions the MCP as a development and testing tool, and with write access it can perform destructive database and project operations, so favor read-only and project-scoped restrictions for casual work.
Redis gives agents live database access through its official MCP server plus coding guidance through official skills. The MCP covers data structures, keys, and query, search, and vector operations with read and write access; the skills carry Redis application patterns.
Use the MCP for operational work against a real instance and the skills for correct Redis application design. Because the server can write, connect it with tightly scoped ACL credentials and prefer read-only or narrow permissions until a task clearly needs more.
Microsoft’s Power BI Modeling MCP for working with semantic models through an agent: inspecting and editing model structure, running DAX queries, and operating across PBIP, Fabric, and Power BI Desktop workflows. It ships as a Public Preview npm package that runs locally.
Use it for supervised semantic-model development, where DAX and model work go through structured tools instead of brittle UI automation. It is Public Preview software with write access that can modify models, so work against backups with limited permissions rather than casually against production models.
A server framework that connects agents to a wide range of SQL, NoSQL, analytics, and cloud databases. It can expose ready-made exploration tools for quick work, or narrowly defined production queries where you control exactly which operations an agent can invoke.
The same server works across Claude, Codex, Cursor, and custom frameworks, handling connections, pooling, auth, and observability in one place. The trade-off worth respecting: prebuilt SQL tools are broad and convenient, but sensitive production data should stay behind read-only accounts and scoped, purpose-built tools.
Google’s warehouse offered through three routes: a managed remote MCP endpoint, a managed Claude connector, and a local path via MCP Toolbox for Databases. Agents can discover datasets and tables, inspect metadata, and run SQL.
Access runs through existing Google Cloud OAuth, IAM, audit, and billing controls, so data stays inside current governance. Prefer the read-only SQL tool. The managed endpoint caps queries at three minutes and 3,000 returned rows and skips Drive external tables, so large pulls need another path.
An official integration that combines a hosted MCP connection with more than 30 skills covering product analytics, feature flags, experiments, error tracking, surveys, dashboards, and LLM analytics.
It gives an agent live product and production context, then lets it investigate trends and errors or manage PostHog workflows without switching tools. Note that the MCP can write, changing flags, experiments, dashboards, and surveys, so grant that access deliberately.
An official integration that takes two forms. The Claude and Codex plugins delegate Snowflake prompts to Cortex Code, while the Cursor plugin connects to a Snowflake-managed MCP server. Both give agents governed access plus Snowflake-specific skills.
Those skills cover SQL, governance, ML, dynamic tables, semantic views, cost, performance, and Cortex services, so an agent works with Snowflake instead of guessing. The two routes share the name but not the same architecture, and queries and Cortex features may consume paid compute.
Chroma’s official MCP server lets an agent create, query, update, and delete vector-database collections and documents, with vector-similarity, full-text, and metadata-filtered search. It runs self-hosted in ephemeral, persistent-local, self-hosted-HTTP, or Chroma Cloud modes.
Reach for it when an agent needs semantic retrieval, a project knowledge base, or explicit persistent memory backed by Chroma. The default ephemeral client loses data on restart, so choose the mode deliberately, and note that collection and document tools can write and delete.
Tableau MCP gives agents governed access to Tableau content and analytics. It comes two ways: a Tableau-managed remote MCP service for eligible Cloud or Server accounts, and an official open-source server you host yourself.
Use it when Tableau already holds your organization’s governed analytics and you want the agent to query it while preserving platform permissions. Prefer the managed MCP when your account SKU supports it; use the self-hosted server for self-managed or unsupported environments.
An official agent stack that pairs ClickHouse-specific skills with live, read-only access to ClickHouse Cloud through MCP. The skills carry current database knowledge; the connection lets an agent inspect real services and data.
Instead of generic SQL advice, an agent can design schemas, optimize queries and ingestion, troubleshoot clients, and inspect actual Cloud services, schemas, costs, and pipelines. The general Cloud MCP is read-only, so it analyzes rather than changes your data.
DuckDB Skills are the project’s official, portable agent instructions for the embedded analytical engine - nine skills covering SQL, performance, data formats, extensions, and workflows. They are guidance, deliberately not a live-database MCP.
Use them for local analytical work where DuckDB is already the execution engine, so an agent writes correct, performant DuckDB SQL without a hosted account. Keep them distinct from MotherDuck’s hosted MCP, which is the cloud service rather than the local engine.
Databricks connects agents to governed lakehouse data and AI assets - Unity Catalog objects, tables, functions, vector search, and Databricks apps - while also shipping an official collection of development skills. Live access runs through Databricks-managed MCP services.
Reach for it for Databricks-centric analytics, data engineering, and AI development, where the agent should work against real catalog data without bypassing governance. Distinguish the live MCP access from the portable skills, which carry development knowledge but do not themselves grant workspace access.
An MCP server built into each enabled Metabase instance. It lets an agent search tables and metrics, inspect fields and sample values, build and run queries, create or update Metabase content, and return interactive charts.
This gives an agent governed analytics context instead of a raw database connection: results and actions stay scoped to the signed-in user’s permissions. Worth knowing that content tools can write, so the same permissions can create or change dashboards, not just read them.
An authenticated product-analytics MCP paired with vendor-authored plugins and skills for analysis, dashboards, experiments, session replay, feedback, taxonomy, and instrumentation, giving an agent governed access to Amplitude projects.
It lets an agent query and edit analytics objects, investigate funnels and experiments, analyze replay and feedback, and plan instrumentation, using repeatable workflows instead of generic prompting. One limit worth remembering: the MCP is not an event-ingestion endpoint, so production events still go through the SDK or HTTP API.
An official toolkit that pairs vector-database tools with guided skills for semantic search, RAG, full-text search, and Pinecone Assistant. An agent can create and inspect indexes, upsert and search records, and rerank results.
This is the integration to reach for when building or operating search, RAG, recommendation, or document-Q&A features on Pinecone, since the agent both follows the workflow and acts on real indexes and data. The MCP centers on integrated-index operations; the CLI and packaged scripts cover broader or batch work.
An official stack that combines analytics-engineering skills with the dbt MCP Server, giving an agent project, semantic-layer, CLI, job, and documentation context for work in a dbt project.
With it, an agent can build and test models, use MetricFlow, troubleshoot jobs, work with Mesh, and run dbt commands using current product guidance. The skills help even without live access, though many assume an existing dbt project, and CLI/SQL tools can change models and warehouse state.
Mixpanel’s official hosted MCP connects an agent to product-analytics work: queries, funnels, retention, dashboards, taxonomy, experiments, feature flags, session replays, and data quality. Claude presents it as a connector; Codex has a Mixpanel plugin.
Reach for it when an agent needs conversational read and write access to Mixpanel product analytics. MCP access must be enabled by an organization admin, and the server can write data and is not covered for HIPAA at research time, so scope it deliberately.
A first-party bundle for moving from a business question and authorized data to a validated analysis, chart, dashboard, notebook, report, or recommendation. The live package currently holds 15 skills, 20 optional apps, and 3 app templates.
Instead of jumping straight to a confident chart, it routes work through business context, data-quality checks, KPI design, metric diagnosis, validation, and visualization. It draws on uploaded files or connected warehouses, but it cannot make weak or mismatched data trustworthy, so validate definitions, joins, and time windows first.
Looker connects agents to governed analytics through Google’s managed MCP service, delivered as an official Claude plugin and as a Preview managed MCP endpoint for other clients. The agent works with Looker’s semantic model and analytics tools rather than a raw database.
Use it for organizations already standardized on Looker that want semantic-model-aware business intelligence - governed metrics and Explores - instead of raw SQL access. It requires admin enablement, and the managed MCP is a Preview service, so confirm its status before relying on it.
MotherDuck’s official integration lets an agent query, prepare, visualize, and manage cloud DuckDB data - reading schemas, running SQL, working with views, and creating shareable Dives. Claude presents it as a connector; Codex has a MotherDuck plugin and app.
Use it when an agent should behave like an iterative analyst over MotherDuck data, moving from schema to SQL to a shareable Dive. It is the hosted cloud service rather than the local DuckDB engine, and queries and writes can change data and incur warehouse usage.
