Key takeaways
- MCP is a shared way for assistants to reach other software.
- An MCP server describes what a tool lets an assistant read and do.
- It works in conversation, not as a scheduled data sync.
- Writes into another tool deserve an approval step.
What is MCP and how does it work
MCP works by pairing a client, the AI assistant, with a server that a software tool provides. The server lists what the assistant may do, such as search records or create a task, and the assistant calls those actions when a conversation needs them.
Access usually goes through the tool's own sign-in, so the assistant acts with the permissions of the person who connected it.
- Client: the assistant that asks for data or actions.
- Server: the tool's published list of actions and data.
- Tools: the individual actions, such as find a contact or update a deal.
- Authorisation: the sign-in that decides what the assistant may touch.
What is an MCP server
An MCP server is the piece of software that exposes a product's actions and data to assistants in the standard format. A vendor, or sometimes a developer, builds and runs it.
For an illustrative CRM, the server might offer actions to search companies, read a contact, create a note and move a deal to a new stage. The assistant sees those actions and their descriptions, and chooses which to call.
How is MCP different from an API or an integration
MCP sits on top of a product's own capabilities and gives every assistant the same way in, while an API is specific to one product and an integration is a fixed connection between two. A typical integration copies data between systems on a schedule; MCP lets an assistant look something up or make one change in the middle of a conversation.
That makes MCP flexible, but it is not a sync. Nothing moves unless an assistant calls an action, so it does not keep two systems matched in the background.
Why MCP matters for go-to-market teams
MCP matters for go-to-market teams because it lets one assistant work across the CRM, billing, scheduling and project tools a team already uses, without copying everything into one place. A seller can ask a question that spans several tools in plain words.
The risk is writes. An assistant that can update a deal or create a record should do so with a person's approval, at least until the team trusts the pattern. In Pebbles Ai, every MCP call follows the execution mode: in Manual it waits for approval, in Plan it is queued, and in Auto it runs without asking.
Where Pebbles Ai fits
- FoundationIntegrationsConnect MCP servers from the Pebbles Ai catalogue, such as Attio, Pipedrive, Stripe and Linear, with each service's own sign-in; the catalogue may change. Pebbles Ai is an MCP client and connects catalogue servers only.
- CapabilityAI workspaceThe Marketing and Strategy assistants use MCP tools in chat, and every MCP call follows the Auto, Plan or Manual execution mode you choose.
Common questions
Who created the Model Context Protocol?
Anthropic introduced MCP as an open standard, and many AI products and software vendors now support it with their own clients and servers.
Is MCP safe to use with a CRM?
It can be, if access uses the CRM's own sign-in and writes need a person's approval. Check which actions the server exposes and start with read-only questions.
Can you add any MCP server to Pebbles Ai?
No. You connect servers from the Pebbles Ai catalogue, each with its own sign-in; adding a server by URL is not supported.
About this page. Written by the Pebbles Ai team for people who run sales, marketing and growth. Last updated .
