Open-source MCP servers make your Data Readiness Hub and Redpoint Interaction data conversational, with any LLM you choose.
Redpoint’s open-source MCP servers connect any LLM directly to your Redpoint data, allowing your own custom agents to look things up, reason about your setup, and deliver insights using real, live data.
The open-source MCP servers, released under the Apache 2.0 license, connect the Data Readiness Hub (DRH) and Redpoint Interaction (RPI) to any LLM via the Model Context Protocol (MCP). Agent developers and engineering teams now have a direct, open connection between whatever LLM they already use and their Redpoint data.
What are Redpoint’s MCP servers?
Redpoint’s MCP servers let you build, connect, and run AI agents against your RPI and DRH instances using any LLM provider you choose. Your teams can ask your DRH and RPI data questions in plain language and get detailed, trustworthy answers.
They wrap a curated set of Redpoint’s Integration APIs as standard MCP tools and route each question to only the tools it needs (keeping token costs down), and ship with a lightweight reference agent so you can see it working out of the box.
Agent developers can clone the repo, study the patterns, and build their own agent or UI layer as it applies to their organization’s setup and needs. With authentication, tool schemas and data access already handled, developers get to spend their time on the fun part: shaping how the agent behaves and what it can offer teams across the organization.
The MCP servers sit on top of your existing Data Readiness Hub or Redpoint Interaction instance, adding a conversational layer that respects your existing RBAC permissions and leaves your data and configuration untouched.
How it works
- Connect. Point any MCP-compatible client, such as Claude Desktop, Cursor, a custom agent you’ve built, or the included reference web agent, at the relevant Redpoint MCP server. Evaluators can run the whole stack with a single `docker compose up`; agent developers can drop in a standalone binary; contributors can clone the repo directly.
- Choose your model. Add the API key for whichever LLM provider you already use. Both MCP servers are built on the Vercel AI SDK’s bring-your-own-model approach, so even if you switch providers, it’s a configuration change rather than a full rewrite.
- Ask, in plain language. The skill router figures out which domain (audiences, interactions, selection rules, data quality) your question falls into and hands the agent only the specific tools it needs, rather than the entire toolset. A foundation layer keeps track of your tenant, client, and ID structure in the background, so you don’t have to feed the agent UUIDs to get a useful answer.
Two MCP servers, one framework
Redpoint currently ships two MCP servers, with more planned as we expand coverage across the platform.
DRH MCP wraps the Data Readiness Hub, giving an agent access to feed status, match data quality trends, and source-level error information. You can get faster triage: instead of digging through the data readiness UI to figure out why a feed failed or why match quality dipped, you can ask the agent directly and get a diagnosis with historical context attached.
RPI MCP wraps the Redpoint Interaction Integration API (audiences, interactions, selection rules, clients, and folders) so an agent can answer questions about your campaigns in natural language instead of requiring someone to open the UI or write a query. Today it’s built for exploration and reporting, tackling some of the investigative questions that tend to crop up frequently across a team, such as “how many audiences ran last month,” “does this folder have duplicate rules in it,” or “how many people currently match this selection rule.” It also carries built-in campaign design guidance that can guide new or infrequent users through selection rule logic and audience design (down to a fully worked win-back campaign example).
What makes Redpoint MCPs different
- LLM-agnostic by design. Both MCPs work with Claude, OpenAI, Azure OpenAI, and more. Swapping providers (or even cloud providers) is made easy with a configuration change.
- Token-efficient by default. The skill router keeps tool exposure narrow, routing a given question to only the 2-3 relevant tools instead of loading the entire catalog. This delivers a meaningful reduction in token overhead compared to handing an agent every available tool at once.
- One setup and governed access. IT and engineering do a single one-time `.env` configuration and security review that every team building on top inherits. The agent also authenticates as the requesting user, so it never sees more than that person’s own access already allows.
- Open source, not a black box. The full repository is available on GitHub under the Apache 2.0 license, so your team can inspect exactly what the agent can and can’t do and extend it if you need capabilities not yet available.
- Compatible with the MCP clients you already use. Because they speak standard MCP, each Redpoint MCP works with Claude Desktop, Cursor, and any other MCP-compatible client without requiring any proprietary plugins.
FAQs
What are the Redpoint MCPs?
They are open-source MCP servers that expose Data Readiness Hub and Redpoint Interaction as MCP servers, so AI agents built on any LLM can query and act on your Redpoint data using natural language.
Which LLMs do they work with?
Redpoint MCPs are LLM-agnostic and work with Claude, OpenAI, Azure OpenAI and others, right out of the box.
Do Redpoint’s MCPs change or modify my existing Data Readiness Hub or RPI instance?
No. They sit on top of your instance and call the same Integration APIs your team already uses They don’t alter your data, your configuration, or your existing permissions.
Who are the Redpoint MCPs built for?
Once agent developers and IT/engineering teams have cloned the repo, wired up the connection and built customized agents or a UI on top of these MCP servers, business teams (analytics, marketing, CX, and other stakeholders) can reap the benefits of natural language queries against their data once agents are running.
Are Redpoint’s MCPs secure enough for a regulated environment?
Redpoint MCPs support authentication, per-user RBAC when connected to your RPI or DRH instance, and audit logging, and are designed to keep the LLM grounded in deterministic tool calls rather than open-ended actions. They are read-only by default, with no delete endpoints today. As with any AI deployment, we recommend reviewing configuration against your own compliance requirements before exposing them beyond a local evaluation.
Where do I get my Redpoint MCP?
Both MCPs are available now on GitHub, with setup docs for evaluators, agent developers, and contributors. You can also find implementation details in the Redpoint documentation.
