Home

How to Use These Docs

Redpanda’s documentation helps you stream data, build pipelines, and deploy AI agents whether you manage infrastructure yourself or run in the cloud. This page explains how our docs are organized, clarifies the relationship between deployment models and products, and shows you how to access documentation through AI-powered tools and programmatic interfaces.

Understand the documentation structure

Redpanda’s documentation is organized around multiple deployment models and products. The Data Platform provides the streaming foundation with Cloud or Self-Managed deployment options. The Agentic Data Plane offers a specialized infrastructure for AI agents and agentic workflows.

Documentation hierarchy
Redpanda Documentation
├── Data Platform
│   ├── Cloud (managed by Redpanda)
│   │   ├── Serverless
│   │   ├── BYOC
│   │   └── Dedicated
│   └── Self-Managed (managed by you)
│       ├── Streaming (broker, Redpanda CLI)
│       └── Connect (data pipelines)
└── Agentic Data Plane
    └── AI agent infrastructure and workflows

What changed: Previously, the software that deployed a Redpanda broker was called "Redpanda Self-Managed". In the new structure, "Self-Managed" refers to the deployment model (you manage the infrastructure), while "Streaming" is the product name for the broker itself.

Key components:

Choose your path

Find the documentation that matches your use case:

I want to deploy Redpanda myself → Self-Managed docs

I want Redpanda managed for me → Cloud docs

I want to integrate data sources/sinks → Connect docs

  • Available in both Cloud and Self-Managed deployments

  • Hundreds of prebuilt connectors including AI and change data capture (CDC)

  • Connect documentation

I want to build AI agent infrastructure → Agentic Data Plane docs

AI agent access

Redpanda’s documentation is optimized for AI agents and LLM consumption. Whether you’re building AI agents, IDE plugins, or automation tools, you can access documentation through interactive page features, a Model Context Protocol (MCP) server for real-time queries, and static markdown exports for offline processing.

Discovery files

Redpanda follows the llms.txt standard for AI agent discovery:

  • https://docs.redpanda.com/llms.txt: AI agent discovery file with curated overview and links to comprehensive resources

  • https://docs.redpanda.com/llms-full.txt: Every component in one file, using the latest version of each

  • Component-specific exports, which are smaller and cheaper to index than the full export: streaming-full.txt, cloud-data-platform-full.txt, connect-full.txt, agentic-data-plane-full.txt, labs-full.txt, and home-full.txt

Older product versions are not in these exports. To get an older version, request the individual page instead (replace .html with .md in any URL).

Page options

Every documentation page includes interactive features:

View as plain text: Click View as plain text in the page options dropdown to render the current page in GitHub-Flavored Markdown format. This option is useful for reading in markdown-aware editors, accessing content offline, or sharing documentation snippets.

Copy as Markdown: Click Copy as Markdown in the page options dropdown to copy the page markdown to your clipboard with YAML frontmatter preserved (includes title, description, git dates, categories, and more). This option is useful for pasting into notes, documentation systems, or issue trackers.

Ask AI: Click Ask AI in the page options dropdown or top-nav section to open an AI chat drawer with full access to Redpanda documentation. The AI can explain configurations in simpler terms, show examples, or highlight common pitfalls. Without signing in, you can ask three questions every 24 hours. Sign in with a free Redpanda Cloud account for unlimited questions and the docs AI agent, which also saves your conversations.

Install the MCP server

Redpanda provides a remote Model Context Protocol (MCP) server that lets you access authoritative Redpanda documentation directly from your IDE or AI tool, such as Claude Code, Cursor, VS Code, ChatGPT, or Claude Desktop.

The MCP server is hosted at: https://docs.redpanda.com/mcp. You can add this endpoint to any AI agent that supports MCP.

Authentication

You can connect and start asking without signing in. Anonymous access is limited to 10 tool calls every 24 hours, counted per network address, so you share that allowance with anyone else using the server from the same network. Connecting, listing tools, and sending feedback don’t count against it: only the other tool calls your agent makes.

Signing in with a Redpanda Cloud account, which is free to create, removes that limit, and gives you an allowance of your own instead of a shared one. There’s no token to copy: the server uses OAuth, and any client that supports the MCP OAuth flow handles sign-in for you.

You sign in once. Your client renews access in the background, and prompts you again only if you go 30 days without using the server.

Your client may not offer to sign you in. MCP clients start their sign-in flow when a server tells them authentication is required, and this server doesn’t, because anonymous access works. So if you want the higher allowance, ask your client to authenticate rather than waiting to be prompted: reconnect the server, or use your client’s own authenticate action for it if it has one.

Signing in to docs.redpanda.com in a browser is not enough on its own. That signs in the browser; your client authenticates separately, with its own OAuth flow. What signing in first does do is make that flow a redirect with nothing to click, which matters for the timing problem below.

If you connect through a stdio bridge such as mcp-remote, sign in to docs.redpanda.com in your default browser before you connect. Your client opens the browser for the sign-in step and waits a limited time for the result, and the host stops the bridge when its startup timeout elapses (30 seconds in Claude Code), taking the browser step with it. Creating a Cloud account takes longer than that. If you’ve already hit this, see Sign-in opens a browser but the page fails to load.

What we collect and why.

When you sign in: we receive your verified email address (and your organization, when available) from Redpanda Cloud, which we use to track documentation usage and attribute it to your organization. This may be shared with our customer and analytics systems and passed to our documentation search provider (Kapa) for usage attribution.

When you use Ask AI without signing in: we count your free questions. Where your cookie choices allow it, we do that with a first-party cookie (__Host-rp_docs_visitor) holding a random identifier: it expires after 24 hours, holds nothing but that identifier, and is not shared with third parties. If you’ve declined that category in Cookie preferences, we store nothing on your device and count against your network address instead, which means you share that day’s free questions with others on the same network.

Either way: so that starting fresh isn’t an unlimited reset, we keep a short-lived count per network address. Those counts are never tied to your questions or to you, and we don’t keep the address itself alongside them.

For details, see our Privacy Policy.

  • Claude Code

  • Cursor

  • VS Code

  • ChatGPT

  • Claude Desktop

Run the following command to add the Redpanda MCP server to Claude Code:

claude mcp add --scope user --transport http redpanda https://docs.redpanda.com/mcp

The server works straight away, on the anonymous allowance. To sign in for the higher allowance, run /mcp in Claude Code and authenticate the redpanda server.

This adds the server as redpanda over the native HTTP transport, for your user account on any platform (stored in ~/.claude.json).

To verify the installation:

claude mcp list

You should see redpanda: https://docs.redpanda.com/mcp (HTTP): ✓ Connected in the output.

For more information about Claude Code, see the Claude Code documentation.

Add the following to your .cursor/mcp.json file:

{
  "mcpServers": {
    "redpanda": {
      "type": "http",
      "url": "https://docs.redpanda.com/mcp"
    }
  }
}

The server works straight away, on the anonymous allowance. To sign in for the higher allowance, reconnect the server from Cursor’s MCP settings.

For more information about MCP in Cursor, see the Cursor documentation.

Prerequisites: VS Code 1.102+ with GitHub Copilot enabled.

Create an mcp.json file in your workspace .vscode folder:

.vscode/mcp.json
{
  "servers": {
    "redpanda": {
      "type": "http",
      "url": "https://docs.redpanda.com/mcp"
    }
  }
}

The server works straight away, on the anonymous allowance. To sign in for the higher allowance, reconnect the server from the MCP view.

To configure globally for all workspaces:

  1. Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P)

  2. Run MCP: Open User Configuration

  3. Add the same JSON configuration

Using MCP in VS Code:

  1. Open the Chat view (Ctrl+Cmd+I / Ctrl+Alt+I).

  2. Select Agent mode from the dropdown.

  3. Click the Tools button to see available MCP tools.

  4. Ask questions about Redpanda and the AI automatically uses the documentation.

For more details, see the VS Code MCP documentation.

ChatGPT connects to remote MCP servers as custom plugins. Account and workspace policies apply, so a workspace admin might need to allow custom MCP servers first.

  1. Go to ChatGPT Plugins.

  2. Select the plus button, then Add custom MCP server.

  3. Enter a name, such as redpanda, and a description.

  4. Under Connection, enter the MCP server URL: https://docs.redpanda.com/mcp

  5. Set authentication to OAuth, review the risk warning, and select I understand and want to continue.

  6. Select Create as a plugin.

The server works straight away, on the anonymous allowance. To sign in for the higher allowance, reconnect the plugin and complete the OAuth flow when prompted.

For more information, see Connect and test your plugin in the OpenAI documentation.

Connect Claude Desktop to Redpanda’s MCP server using one of two methods:

Method 1: Using Connectors

This method works on all plans and all platforms (macOS, Windows, Linux). The Free plan is limited to one custom connector. On Team and Enterprise plans, an organization owner adds the connector first, and members then connect to it.

  1. Open Claude Desktop.

  2. Go to Customize > Connectors.

  3. Click + Add, then Add custom connector.

  4. Enter a name, such as redpanda, and the URL https://docs.redpanda.com/mcp, then click Continue.

  5. Review the detected authentication settings, then click Add.

  6. The connector works straight away, on the anonymous allowance. To sign in for the higher allowance, reconnect it and complete the OAuth flow when prompted.

When using Connectors, Claude connects to your remote MCP server from Anthropic’s cloud infrastructure.

Method 2: Using Configuration File

This method works for all plans, including Free plan.

Edit the Claude Desktop configuration file for your platform:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following configuration to your claude_desktop_config.json file:

{
  "mcpServers": {
    "redpanda": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://docs.redpanda.com/mcp"]
    }
  }
}

This configuration uses the mcp-remote bridge to connect to Redpanda’s remote MCP server. On first use, mcp-remote opens a browser for you to sign in with your Redpanda Cloud account. Claude Desktop supports only stdio and sse transports, so mcp-remote converts the HTTP endpoint to a compatible format.

Before you restart, sign in at docs.redpanda.com in your default browser. Claude Desktop stops mcp-remote when its startup timeout elapses, so a sign-in that involves creating an account cannot finish in time. See Authentication.

Restart Claude Desktop for changes to take effect.

On Linux, Claude Desktop cannot connect to remote servers configured using claude_desktop_config.json. If this method doesn’t work, use the Connectors interface instead.

For more details, see the Claude custom connectors documentation.

Other AI tools

Any tool that supports the MCP OAuth flow can connect using the following URL and will prompt you to sign in with your Redpanda Cloud account on first use:

https://docs.redpanda.com/mcp
MCP support varies by tool and version. Check the specific tool’s documentation for MCP setup instructions. See Authentication for details on sign-in.

What you can do

Once connected, you can ask context-aware questions about Redpanda from within your editor:

  • "How do I configure Redpanda for production?"

  • "What are the instructions for running Redpanda in a local Kind cluster?"

  • "What are the best practices for topic partitioning in Redpanda?"

  • "What are Redpanda’s security features and authentication methods?"

  • "How do I monitor Redpanda cluster performance?"

  • "What’s the difference between Redpanda Cloud and self-hosted deployment?"

For facts with one right answer, the server also has exact lookups that read the generated reference data instead of searching prose: a configuration property by name (its type, default, whether it needs a restart, whether Redpanda Cloud supports it), a search across properties by keyword or scope, and a Redpanda Connect component by type and name (its status, fields, and defaults). Your assistant picks these when you ask things like "What is the default of log_segment_size in 26.1?", "Which properties control retention?", or "Does the redpanda output support batching?". A property or component that does not exist comes back as not found, with close matches, rather than a guess.

You can also send feedback to the Redpanda team through your AI client. If you hit a bug, a documentation gap, or anything unclear, tell your assistant something like "send feedback that this page is missing X" and it submits it for us to review. Your assistant may also report problems it notices on its own. Feedback doesn’t include your name or email address, so don’t put personal details in it. To hear back from us, contact Redpanda support.

Usage limits

To ensure fair use and performance for all users, the MCP endpoint enforces the following rate limits:

  • 60 requests per 15 minutes

When you’re signed in, this limit is applied per user. Unauthenticated requests (during the rollout period) are limited per IP address.

As well as this limit, the ask_redpanda_question tool proxies to an MCP server hosted by Kapa.ai, which enforces its own limits. See the Kapa documentation for details.

These limits are suitable for:

  • Individual developer IDE usage

  • Ad-hoc documentation queries

  • Evaluation and testing

If you exceed the limit, you receive an HTTP 429 response with rate limit headers. Wait until the reset time before retrying.

Rate limit exceeded response
HTTP 429 Too Many Requests
RateLimit-Limit: 60
RateLimit-Remaining: 0
RateLimit-Reset: <timestamp>

Troubleshooting

Authentication required (HTTP 401)

Anonymous access is allowed, so a 401 here means your client sent credentials the server could not accept, rather than none at all. If you receive an HTTP 401 response with an authentication_required error:

  • Reconnect or restart the client. If sign-in succeeded before, your token may have expired, and reconnecting starts a fresh sign-in.

  • Check that your client supports the MCP OAuth flow (most recent versions do). A client that doesn’t can still use the server anonymously.

  • If you don’t have a Redpanda Cloud account, create a free one.

Out of free tool calls

If a tool returns an anonymous_quota_exhausted error, you’ve used the anonymous allowance of 10 tool calls in a 24-hour window. Your agent usually reads the message out rather than showing the error, so it may reach you as a sentence about signing in.

The allowance is counted per network address, so on a shared or office network you may reach it without having made 10 calls yourself. It resets 24 hours after the first call it counted.

To get an allowance of your own, sign your client in. Reconnect the server, or use your client’s authenticate action for it: your client won’t prompt you on its own, because anonymous access works and nothing tells it authentication is required. See Authentication.

Sign-in opens a browser but the page fails to load

If the browser ends up on a localhost address reporting ERR_CONNECTION_REFUSED, if sign-in keeps opening a new tab without ever finishing, or if the browser shows a Finish connecting button after you sign in, your MCP client stopped listening for the result before you finished in the browser. Clients that connect through a stdio bridge such as mcp-remote are affected: the host stops the bridge when its startup timeout elapses (30 seconds in Claude Code), and the bridge’s browser listener dies with it. The --auth-timeout option of mcp-remote does not extend this window.

To get through it:

  • Sign in to docs.redpanda.com in your default browser first, then connect your MCP client. The client’s sign-in step is then a redirect with nothing to click, which finishes in seconds.

  • If the browser shows Finish connecting, click it within a minute. If nothing happens, reconnect the server in your client.

  • If you run mcp-remote under Claude Code, raise the startup timeout by setting MCP_TIMEOUT=300000 in the environment Claude Code runs in, and clear any half-finished state with rm -rf ~/.mcp-auth before retrying.

  • For Claude Code, Cursor, and VS Code, use the HTTP transport shown in the setup section instead of mcp-remote. Those clients keep listening for as long as the session is open, so the window is not a factor.

If you were creating your Redpanda Cloud account when this happened, no timeout setting is long enough to cover it: signup includes a verification email and Cloud onboarding. Finish creating the account in the browser, then reconnect your client. That sign-in is a redirect with nothing to click, because you’re already signed in to Redpanda Cloud in that browser.

Account setup is not finished

If sign-in fails with access_denied and a message about finishing account setup, your Redpanda Cloud account exists but does not belong to an organization yet. Verifying the address in the Cloud signup email does not create one. Complete the Redpanda Cloud onboarding steps, then reconnect your client. Retrying sign-in before onboarding is finished fails the same way every time.

Work account required (HTTP 403)

Any verified Redpanda Cloud account works by default, including personal email providers. If your organization has opted to restrict access to work email domains and you receive an HTTP 403 with a work_email_required error, sign in with your work account instead.

Server not connecting

  • Verify the URL is exactly: https://docs.redpanda.com/mcp.

  • Check your internet connection.

  • Ensure your AI tool supports HTTP-based MCP servers.

  • Restart your AI tool after adding the configuration.

VS Code specific issues

  • Ensure you have VS Code 1.102 or later.

  • Verify GitHub Copilot is installed and enabled.

  • Try running MCP: Reset Cached Tools from the Command Palette.

  • Check the Output panel (View > Output > MCP) for error messages.

Configuration issues

  • For Claude Code, Cursor, and VS Code, use the HTTP transport (--transport http or "type": "http").

  • For Claude Desktop, use the mcp-remote bridge shown in the setup section. Claude Desktop does not support direct HTTP transport.

  • If you previously configured Claude Code using mcp-remote and experience connection issues, remove the old configuration and use the HTTP transport method shown above.

  • Some MCP clients may have issues with SSE streaming. If you experience connection problems, verify that your client supports HTTP-based MCP servers with Server-Sent Events (SSE).

  • Check that your client’s Accept headers include both application/json and text/event-stream.

Rate limiting

If you’re building automation or high-volume integrations and hitting rate limits frequently:

  • The public endpoint’s rate limits are intentionally restrictive for fair use.

  • Consider implementing caching on your side to reduce duplicate queries.

Other issues

Check the MCP GitHub repository for additional troubleshooting guidance, or contact Redpanda support to report a problem with our documentation.

Static markdown exports

For agents without MCP support or for offline processing, use our static markdown exports:

Component-specific exports:

Use cases:

  • Offline AI training and model fine-tuning

  • Batch processing and analysis

  • Full-text search implementations

  • Custom documentation tools

Fetch individual pages

Every HTML page in our documentation has a markdown equivalent that you can access programmatically.

  1. Direct access: Replace .html with .md in any URL

    Example: https://docs.redpanda.com/streaming/current/get-started.html → https://docs.redpanda.com/streaming/current/get-started.md

  2. Content negotiation: Request any URL with Accept: text/markdown header

Markdown features:

  • YAML frontmatter with page metadata (title, description, categories, topic type, personas)

  • Git dates (accurate created/modified dates from commit history)

  • GitHub-Flavored Markdown format

Use cases:

  • API integrations

  • Automated documentation pipelines

  • Custom documentation workflows

  • Integration with internal knowledge bases

Contributing:

  • Found a typo or have a suggestion? Open an issue or pull request on GitHub