Why software companies are racing to ship MCP servers

Software companies are shipping MCP servers because their buyers are starting to run AI agents next to their human teams. Those agents work in whichever AI client or framework the buyer chooses, and they use the tools they can reach. A product that agents cannot use gets routed around. An MCP server is how a product stays in the workflow.

Key takeaways

• MCP gives any MCP-capable agent one standard way to discover and use a product. Anthropic counted more than 10,000 active public MCP servers in December 2025.
• HubSpot's MCP server gives agents read and write access to CRM records, activities, content and campaigns.
• Brand Influence, an influencer marketing platform, built its workflow so agents can run campaign work through MCP.
• Buyers should add "does it have an MCP server, and what can it do" to every software checklist.
• Vendors should ship read tools first, then write tools with approval steps, clear descriptions and scoped permissions.

If MCP is new to you, start with our explainer, What is MCP?

Why does a tool without MCP get routed around?

Until recently, a person sat between every piece of business software. They opened the CRM, copied a number into a spreadsheet and pasted a summary into an email. Agents change that. A team can now ask an assistant to pull the pipeline, draft the follow-ups and update the records, and the assistant does the steps itself.

That only works with tools the agent can reach. If the CRM has an MCP server and the project tool does not, the agent will happily keep its notes in the CRM, or in a document, and the project tool slowly stops being the place where work is tracked. Nobody decides to drop it. It just stops being used.

Buyers also want a choice of agents. Anthropic says MCP has been adopted by ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code, and OpenAI's API can connect models to remote MCP servers. A vendor that ships one MCP server works across all of them, instead of building and maintaining a separate plugin for each assistant.

Case 1: HubSpot

HubSpot offers two MCP servers. Its developer site says the remote HubSpot MCP server gives "any MCP-compatible AI tool or agent secure read and write access to your HubSpot CRM data," while a separate local Developer MCP server helps developers build on the HubSpot platform. HubSpot's docs describe the remote server as "a bridge between AI systems and HubSpot's APIs."

According to HubSpot's documentation, an agent can:

• Read CRM records, including contacts, companies, deals, tickets, leads and custom objects; activities such as calls, emails, meetings, notes and tasks; blog posts, landing pages, site pages, campaigns and marketing events; and content analytics such as views and conversion rates.
• Write by creating and editing contacts, companies, deals, tickets and other CRM records, logging activities, creating, editing and publishing pages and blog posts, updating campaigns, creating marketing email drafts and managing pipelines.

Two details matter for buyers. HubSpot says the server respects conditional field logic, including custom pipeline stage validations, so an agent cannot skip rules your team set. And tool availability depends on your subscription, user permissions and account setup, so the agent sees only what that user is allowed to do.

Case 2: Brand Influence

Brand Influence is a Stockholm influencer marketing company that runs the influencer workflow in one platform, with clients and campaigns organised by market and a booking pipeline for creators. It was built agent-first: its MCP server lets Claude, ChatGPT or another MCP-capable assistant work as a campaign manager or analyst. Through it, an agent can:

• search creators
• manage the booking pipeline, from outreach to negotiation to booked
• record agreed terms
• schedule follow-ups
• pull evidenced campaign reports

Influencer campaigns involve a lot of small, repetitive steps across many creators, which is exactly the kind of work agents handle well when the platform lets them. The point here is not that every vendor needs an agent-first product. It is that once agents can do the routine steps inside a platform, the people using it spend their time on the decisions.

What does this mean for buyers?

Add one question to your software checklist: does it have an MCP server, and what can it do? Then dig into the answer:

1. Is the server official, built and maintained by the vendor?
2. Which AI clients does it work in, such as Claude, ChatGPT or Copilot Studio?
3. Which tools read data and which write, send or delete?
4. Which permissions does it ask for, and can you narrow them?
5. Can write actions require a person to approve them?

We have added this question to our AI vendor security and procurement checklist. To compare tools, use the Has MCP server filter in our directory, or narrow it down with Works with Claude and Works with ChatGPT. Tools with official servers include Notion and Linear.

What does this mean for vendors?

1. Ship read tools first. Search and reporting carry the least risk and let customers build trust.
2. Add write tools with approval steps. The MCP specification says there should always be a human in the loop who can deny tool calls. Design write tools so the client can show a clear confirmation.
3. Write good tool descriptions. The agent chooses tools from their names and descriptions. Vague descriptions lead to wrong calls.
4. Scope permissions narrowly. The MCP security guidance warns against requesting every scope up front. Ask for the minimum and expand when needed.
5. Respect your own business rules. Like HubSpot's pipeline validations, the server should enforce the same rules as your app.
6. Get listed. Submit to the Claude connectors directory and the ChatGPT apps directory, and document the server on your own site.

Bottom line

Agents are becoming a second set of users for business software, and they use the tools they can reach. MCP is how a product stays reachable, which is why vendors from HubSpot to smaller specialists are shipping servers. Buyers should ask about MCP in every evaluation. Vendors should start with read tools and earn the right to write.