Inflection.io Releases v2 of B2B Marketing Automation MCP to be Agent First Alternative to Marketo
.png)
A year ago we shipped the first MCP server built for B2B marketing. Today we've shipped v2.
We bet on MCP servers a year ago, because we believe it's the interface that matters: a standard way for AI models, agents, and other MCP servers to work together. That bet is paying off. Our customers are running real work through the MCP today, building audiences, launching campaigns, and analyzing performance via Claude, ChatGPT, Slack, and Claude Code in VS Code instead of our UI.
Our philosophy with AI is simple: meet you where you are on your AI journey. If your company hasn't adopted ChatGPT or Claude yet, use our in-app chat or Slack integration, which talks to the same MCP server and gives you the same agentic capability. If your team already lives in Claude or ChatGPT, connect the MCP server and keep working there. If you're AI-native and building in Claude Code CLI, the MCP server is available there, too.
Today we’re releasing 30 new capabilities to the MCP. It can now do everything a human can do to build and run a campaign. Build audiences, draft personalized emails, and construct journeys.
Under the hood, v2 of the MCP server runs on Fabric, the agent harness behind every Inflection agent. Instead of hardcoded workflows, each agent runs off a playbook: a documented recipe of steps, clarifying questions, and guardrails. Customizing an agent to fit how your team actually works means just editing a playbook.
Fabric's other unlock is agent parallelization. The MCP can now take a campaign brief, whether it’s a Notion doc, an Asana ticket, a Google Doc, wherever the audience, offer, and timing are already spelled out, and build the entire campaign from it. An agent reads the brief, breaks the campaign into its component pieces, and spins up parallel agents to build each one at once: one on the audience, others drafting every email in the sequence, etc. They converge into a single journey, branches, timing, audience, and copy assembled together, done. That collapses a handoff that used to cost days. Agents that execute don't just answer questions about your campaign. They build it, in parallel, from the brief you already wrote. Reading from Notion, Slack, Asana, etc.
All of this is what makes our Forward Deployed Operator model special. We put a human operator inside your team to build the specific agents you need, wired into how your team already works. These operators are all former marketing operations professionals, so they understand your challenges and offer suggestions. For one customer, that meant an audience agent that reads and writes directly into their Asana, because that's where campaign build tickets existed.
The last piece is skills: Inflection now has publishable, reusable skills. A skill is your team's process written down once. How you build a newsletter, your QC checklist before a send, your brand voice and guardrails. This is perfect for newsletters, product releases, and other consistent batch campaigns.
Finally, AI is only as good as the context it's given. That's the problem with most MCP servers: they connect a model to a system, but the data underneath is thin, siloed, or stripped of the signal that actually matters. Inflection’s is backed by ContextGraphTM, which maps how people, accounts, product usage, CRM data, website behavior, and marketing activity connect to each other. Not just what happened, but how it relates. That structure is what lets AI reason over your data instead of guessing at it. Full fidelity in, full fidelity out: the best context in the industry produces the best agent output in the industry.
Take a look at the new MCP by requesting a demo https://www.inflection.io/demo
