Best MCP servers: a curated MCP server list

The best MCP servers are the ones that supply the missing tools and context for your actual task, with an access boundary you can explain. This curated MCP server list groups supported AIPx integrations by job, deployment, authentication and cost. Entries were checked against source documentation and the product catalog on October 14, 2026; this is not a claim that every provider was tested end to end.

How we picked these MCP servers

We selected services already supported by AIPx and useful to an individual developer or personal agent workflow. Some entries connect official MCP endpoints; others expose a supported REST API or custom integration as MCP tools. That difference matters when comparing them with an independently installed vendor server.

The selection considers the information a server adds, the operations it exposes and the account access required. A tracker supplies requirements. An error system supplies failure evidence. Mail and meeting tools supply context that is often private. Payment and infrastructure tools deserve a deliberate action boundary rather than blanket availability.

Five checks are useful before connecting anything:

1. Identify the task and the information absent from the current conversation or repository.
2. Check the maintainer and actual server or API source.
3. Match the transport and authentication to the client you use.
4. Review the available operations and account privileges.
5. Try one narrow read, then inspect what account and tool produced the result.

“Official” describes the upstream source, not a guarantee that all data or actions are safe for every agent. “Read-only” describes effects on the source system, not whether its information is public. Prefer a small set that supports the workflow over a list chosen by tool count.

How to read deployment, auth and price

All integrations in the tables below are reached remotely through AIPx. The upstream column distinguishes an official MCP connection from a REST or custom adapter. It does not mean you must run those adapters locally. Local reference-server alternatives are discussed later.

Authentication describes the connected service account, separately from the client’s OAuth login to AIPx. The price column distinguishes AIPx access from the provider’s own plan. AIPx Free includes three connected accounts and one agent; Pro costs $3 per month with unlimited accounts and agents plus Pro integrations. Service subscriptions, API entitlements and rate limits remain separate.

MCP servers by category

Code, issues and production evidence

Integration	What it adds	Deployment and upstream	Account auth	Cost boundary
GitHub	Repositories, issues, pull requests and Actions	Remote, official MCP	OAuth	AIPx plan plus GitHub account limits
Linear	Issues and project workflow context	Remote, official MCP	OAuth	AIPx plan plus Linear plan
Sentry	Error investigation and release context	Remote, official MCP	OAuth	AIPx plan plus Sentry plan
PostHog	Product analytics context	Remote, official MCP	Supported account connection	AIPx plan plus PostHog usage
Jira	Ticket requirements and issue operations	Remote, supported REST adapter	Atlassian account API token	AIPx plan plus Jira access
Confluence	Shared technical documentation	Remote, supported REST adapter	Atlassian account API token	AIPx plan plus Confluence access
Atlassian Rovo	Atlassian’s remote MCP interface	Remote, official MCP	OAuth	AIPx plan plus Atlassian entitlements
Supabase	Database and project tools	Remote, official MCP	OAuth	AIPx plan plus Supabase usage
Cloudflare	Developer infrastructure context	Remote, official MCP	OAuth	AIPx plan plus Cloudflare usage

GitHub is the first choice when the missing context lives in issues, reviews or remote repository metadata. It adds something beyond local Git, but that is not a reason to enable every write operation. Choose relevant tool groups and inspect the requested scopes; the connected interface is not a promise of every GitHub administrative capability.

Linear fits issue-driven development. Read the live brief and acceptance criteria before changing code, then use a reviewed update to communicate the result. Keep the issue’s authority distinct from instructions embedded in unrelated comments or retrieved content.

Sentry supplies the evidence a checkout usually lacks: the production failure and its diagnostic context. Start with the application’s project and a named issue. Error payloads can contain private data, so broad read access still deserves review.

PostHog is useful when the question is how a feature behaves in use rather than whether the code compiles. Combine it with error and issue context to avoid making a product decision from one signal alone. Inspect the current account’s tool inventory instead of assuming every analytics feature is available through MCP.

Jira is a good fit when tickets hold the requirements. The supported REST integration is distinct from Atlassian Rovo’s MCP endpoint. Choose the route according to account setup and needed operations, not just because both mention Atlassian.

Confluence helps recover the design decisions behind a ticket. It pairs naturally with Jira, but page access is determined by the connected account and upstream spaces. Hide editing operations when the job is simply reading architecture context.

Atlassian Rovo is the official remote MCP route for supported Atlassian workflows. It is useful if that connection model fits your organization. Do not assume it is interchangeable with the separate Jira and Confluence REST adapters or that it grants every operation from their full APIs.

Supabase adds database and project context to application work. Prefer project-scoped and read-only options where appropriate, then restrict unnecessary tools. Database inspection can expose customer information even when no query modifies rows.

Cloudflare fits infrastructure research alongside code changes. Connect only the account and administrative capabilities the task needs. Production changes deserve a separate authorized deployment workflow; discovering an infrastructure tool should not automatically authorize its use.

Google accounts and planning data

Integration	What it adds	Deployment and upstream	Account auth	Cost boundary
Gmail	Mail search, messages, drafts and labels	Remote, supported Google API	Google OAuth	AIPx plan plus Google account restrictions
Google Calendar	Calendar events and scheduling context	Remote, supported Google API	Google OAuth	AIPx plan plus Google account restrictions
Google Sheets	Spreadsheet data and updates	Remote, supported Google API	Google OAuth	AIPx plan plus Google API limits
Google Maps	Places, routes and geographic data	Remote, supported API	Catalog connection requirements	Provider API costs and access rules apply

Gmail is useful for a specific work conversation or inbox task. Distinguish reading, drafting and sending in both the prompt and tool rules. A personal mailbox should not become available to a coding assistant merely because the work mailbox is useful.

Google Calendar supplies event context for scheduling and preparation. Keep the intended calendar explicit when several accounts are connected. A read-only event view can still reveal private attendees, locations and meeting descriptions.

Google Sheets fits operational tables, test matrices and lightweight planning data. Begin with a small explicit range rather than requesting an entire workbook. Changes to an existing sheet should identify the exact sheet and cells before you approve them in a client that requests review.

Google Maps fits place and route questions that require structured geographic data. Check the provider’s OAuth access and request limits before connecting. This is not a promise that a gateway subscription includes every Maps request or every feature in the consumer map application.

Notes, meetings and personal tasks

Integration	What it adds	Deployment and upstream	Account auth	Cost boundary
Notion	Workspace pages and knowledge	Remote, official MCP	OAuth	AIPx plan plus Notion access
Granola	Meeting notes and related context	Remote, official MCP	OAuth	AIPx plan plus Granola account access
Fireflies	Meeting transcripts and notes	Remote, official MCP	OAuth	AIPx plan plus Fireflies account access
Things 3	A supported personal task workflow	Remote, custom integration	Catalog-specific setup	AIPx plan plus Things requirements
Apple Calendar	Apple calendar access	Remote, custom integration	Supported account-specific credentials	AIPx plan plus Apple account requirements

Notion is useful when product decisions live in workspace pages. Confirm that the connected workspace contains the relevant material and avoid assuming the agent sees every page you can see personally. Shared document edits should identify their target precisely.

Granola helps when meeting notes contain decisions missing from the tracker. Its remote MCP connection uses OAuth. Treat meeting content as private context and verify that the agent’s job actually requires that meeting before exposing it.

Fireflies is another meeting-context option, especially when transcripts are the source for follow-up work. It uses its own account and remote connection rather than borrowing Granola’s setup. Choose the service where the meeting actually lives instead of connecting both by default.

Things 3 serves a personal task workflow through AIPx’s supported custom integration. Read the connection page for its concrete setup and capabilities before assuming it is a general cloud task API. A title saying “MCP” does not erase the underlying app’s operating model.

Apple Calendar has its own account and credential setup. It should not be configured using the Google Calendar OAuth instructions. Connect the intended account separately and review which event information the agent should receive.

Payments and billing

Integration	What it adds	Deployment and upstream	Account auth	Cost boundary
Stripe	Billing and payment-system tools	Remote, official MCP	Supported Stripe MCP account flow	AIPx plan plus Stripe account restrictions
Paddle	Subscription and transaction context	Remote, supported API integration	Catalog credential setup	AIPx plan plus Paddle account restrictions

Stripe can supply billing context for implementation or support questions. Financial writes are a different authority level from reading a payment record. Inspect the tool inventory and account privileges before offering those operations to a general-purpose agent.

Paddle is useful when that is where subscriptions and transactions are managed. Its setup is not inherited from Stripe merely because both are payment providers. Keep the merchant account and requested action explicit, particularly when debugging billing alongside code.

Health and training data

Integration	What it adds	Deployment and upstream	Account auth	Cost boundary
WHOOP	Recovery, sleep and training context	Remote, supported API	OAuth	AIPx plan plus WHOOP membership/access
Withings	Connected health measurements	Remote, custom API integration	OAuth	AIPx plan plus Withings access
Hevy	Workouts, routines and activity history	Remote, supported API	API key	AIPx plan plus Hevy API entitlement

WHOOP is appropriate for an agent explicitly tasked with interpreting your training context. It is not a substitute for medical advice or a reason to expose health information to a coding project. Keep that account within a personal agent boundary.

Withings supplies measurements from its connected account. Treat it as a separate data source, with its own authorization and meaning, rather than a synonym for WHOOP. A useful workflow asks for a defined period or measurement instead of indiscriminate history.

Hevy focuses on workouts and routines. Its API key and provider entitlement are distinct from OAuth-based health connections. Limit access to the task and avoid interpreting a training record as a medical diagnosis.

Search and external research

Integration	What it adds	Deployment and upstream	Account auth	Cost boundary
Exa	Web search and content retrieval	Remote, official MCP	No account credential for the catalog endpoint	Pro integration; provider endpoint limits apply

Exa is useful when a task needs current external sources rather than private account records. The catalog connection is a Pro integration and does not require an account credential for its configured endpoint. Retrieved pages remain untrusted data; search results do not authorize follow-up actions in other connected systems.

Additional infrastructure, inbox and account data

Integration	What it adds	Deployment and upstream	Account auth	Cost boundary
DigitalOcean	Selected cloud services, Droplets and billing context	Remote, official MCP server family	OAuth with selected services and tool groups	AIPx plan plus DigitalOcean resource costs
AgentMail	Mail operations for one configured inbox	Remote, AIPx native API adapter	API key; prefer inbox-scoped access	AIPx plan plus AgentMail usage and account requirements
Numon	Food diary, nutrition reports and fitness profile	Remote, Numon MCP connection	OAuth	AIPx plan plus provider terms; no fixed Numon price verified
X	Post search and authorized account activity	Remote, official X MCP	User-context OAuth	AIPx Pro required; X API entitlements and costs separate
Yandex Direct	Advertising campaigns, ads, bids and reports	Remote, custom adapter over the official Direct API	OAuth; advertiser API approval required	AIPx plan plus provider API units and restrictions

DigitalOcean fits a task that needs the selected account’s infrastructure context. Choose services and tool groups rather than assuming every cloud capability is exposed. Start with a known Droplet or billing record and hide administrative operations irrelevant to the investigation.

AgentMail is an inbox-specific connection, not a proxy for every feature of the broader vendor MCP server. All its tools start on Hide in AIPx, including reads. Explicitly expose only the needed inbox operations, and verify client review behavior before using outgoing mail.

Numon supplies nutrition and fitness context through the connected account. The verified AIPx metadata includes diary, profile and date-range report reads; do not infer that every write from the separate Numon application is available. Nutritional calculations are estimates and should stay within a personal-data boundary.

X is a Pro integration. Available account operations depend on the authenticated upstream tool inventory and X entitlements; a gateway subscription does not include unrestricted API usage. Publishing and account mutations need deliberate review, while irrelevant capabilities should remain hidden.

Yandex Direct uses a supported API adapter rather than an asserted official vendor MCP server. Campaign, ad and bid changes can affect advertising activity, so separate reporting from mutations. The provider’s advertiser access, API units and agency-account rules remain in force.

The wider integration catalog contains supported services beyond these editorial selections. A service without a published page can still appear in the dashboard catalog, but this list does not promise an unsupported connector or substitute a guessed tool inventory.

Best MCP servers for Claude Code

For issue-driven development, begin with GitHub, Linear or Jira, then Sentry for failure evidence. Add Confluence or Notion if architectural decisions are there. PostHog is useful when the question includes feature usage, while Supabase or Cloudflare adds infrastructure context only when needed.

A ten-service shortlist for development is GitHub, Linear, Sentry, PostHog, Jira, Confluence, Notion, Supabase, Cloudflare and Google Sheets. Stripe can be an eleventh choice for a billing project. These are candidates, not a recommended default enabled set.

Claude Code already has local file and shell tools. Installing another broad filesystem server is unnecessary unless it supplies a boundary or environment the built-in tools do not cover. Follow Claude Code MCP setup for HTTP registration, configuration scope and OAuth.

Best MCP servers for Cursor

Cursor benefits from the same external evidence but has its own project and global JSON configuration. Start with GitHub, the active tracker and Sentry. Add the document source behind the requirements, then only the database, analytics or infrastructure connection needed for the specific change.

A useful editor shortlist is GitHub, Linear, Sentry, Supabase, PostHog, Cloudflare, Jira, Confluence, Notion and Google Sheets. Keep payments and personal mail out of a general repository workflow unless the task requires them. Cursor MCP setup explains server registration and tool toggles.

Client-side selection reduces irrelevant choices, but server-side Hide supplies the enforced access removal in AIPx. Allow and Confirm communicate execution hints. Confirm requests client review; actual approval depends on that client and must not be treated as a universal gateway block.

MCP registry, marketplace and directories

A registry helps publish and discover server metadata. A directory helps people browse choices. A marketplace may add installation, hosting or commercial packaging. A listed server still needs a maintainer, a trustworthy endpoint and suitable access controls.

The official MCP Registry is a discovery source. Its presence is not a blanket security endorsement for every published operation or deployment. Follow a listing to its maintained source and documentation, then evaluate the actual endpoint or package you will use.

For a directory candidate, check whether it is still maintained, whether a hosted endpoint exists and whether the instructions match your client. Avoid copying credentials from examples or installing an old package just because its listing ranks highly. A stale setup snippet can remain online after the official server has changed.

Awesome MCP servers on GitHub

Community “awesome” lists are useful leads, but their popularity does not validate account permissions or every package in the list. Review each project’s current maintainer and installation instructions separately.

The MCP reference-server repository now emphasizes a small set of educational reference implementations and points server discovery toward the registry. Some older examples are archived. A reference filesystem or memory server can teach the protocol, but the repository explicitly asks developers to assess production requirements.

For local files, choose a narrow allowed directory rather than the whole home folder. For code and shell work, check what the client already provides. For persistent agent memory, compare a local memory model with AIPx’s built-in KV storage before assuming that every “memory server” behaves the same way.

How to add any of these to your agent

1. Pick the task and one necessary service.
2. Connect the account through its supported method.
3. Create the agent and review the exposed tools.
4. Copy the agent endpoint and authenticate the client.
5. Run a narrow read and inspect the account, tool and result in the audit.
6. Add further operations only when the task requires them.

Use the client-specific guide: Claude, Claude Code, Codex, Cursor, Claude Desktop, ChatGPT or VS Code and other clients. A shared server URL does not make their configuration formats interchangeable.

A direct vendor endpoint may be simpler for one service. Local or privately operated servers fit private environments. AIPx adds another hosted operator in exchange for shared connected accounts, scoped agents and an audit. Compare the gateway model and remote server options before choosing. Connect free.