Key takeaways:
- Salesloft’s MCP server is strictly read-only. It supports no updates, writes or deletes, and cannot search or retrieve cadences or cadence performance.
- It ships 13 tools covering accounts, people, opportunities, conversations, teams and users, which is a deliberately narrow catalog.
- Access is an add-on for Agentic customers, meaning teams that bought an Agent package such as Account Agent or Coaching Agent.
- Salesloft’s own help center describes the server as being in a Crawl phase, and its supported clients are Claude Desktop and ChatGPT.
Salesloft has taken the most conservative approach to MCP of any vendor in this review, and that is a defensible position rather than a shortcoming.
The server reads revenue data and does nothing else. It cannot change a record, advance a deal or touch a cadence, so the worst outcome of a misfiring prompt is a wrong answer rather than a wrong action.
The complication is that Salesloft’s marketing and Salesloft’s documentation currently describe different products. This review works from the help center, which is the surface a customer can actually verify today, and flags where the announcements point somewhere else. We compete with Salesloft and publish our own MCP server, which is worth knowing as you read the comparison.
What Salesloft MCP Is
Salesloft MCP is a standardized interface built on the Model Context Protocol that lets external AI assistants query your Salesloft data. Salesloft’s own FAQ describes it as currently in the Crawl phase, which is the company’s own signal about maturity.
The security posture is the strongest part of the documentation, and it is unusually specific. Data comes from your own Salesloft instance, nothing is cached outside the Salesloft perimeter, and the assistant only sees data during the active chat session. MCP falls inside Salesloft’s existing SOC 2 and ISO 27001 scopes.
One caveat there is stated plainly and deserves repeating, because it is the honest version of a claim most vendors fudge. Salesloft does not use your data for training, but whether the LLM provider does depends on the contract you hold with that provider. The server cannot control what happens once data reaches Claude or ChatGPT, and Salesloft says so.
The Clari Context
Salesloft merged with Clari, and the combined company positions the MCP server as a route into the full revenue lifecycle rather than just sales engagement. Its September 2026 announcement describes adding Cadence and activity data, Clari Copilot call intelligence, forecasting and deal inspection data, plus connectors for ChatGPT, Microsoft Copilot and Google Gemini.
That is roadmap, not current documentation, and the gap between the two is the single most important thing to understand before planning around this server. More on that below.
The Agent packages that gate access are worth understanding on their own terms, because they are the reason most teams will or will not have this server. Account Agent and Coaching Agent are autonomous products rather than features, priced separately, and the MCP server arrives as an inclusion with them. If the category is unfamiliar, our explainer on what an AI SDR is covers how agents differ from the automation already inside these platforms.
What Salesloft MCP Can Do
The catalog is 13 tools, and it splits cleanly into direct lookups and searches. Everything here reads.
| Tool | Object | What it returns |
search_accounts |
Accounts | Salesloft ID, account name, owner, plus account and custom fields |
account_by_id |
Accounts | Company data including name, description and domain, with references to owning and creating users |
search_people |
People | ID, first and last name, email, job title |
person_by_id |
People | Contact or lead records stored in Salesloft |
search_opportunities |
Opportunities | Opportunity name, type, amount, stage, account name |
opportunity_by_id |
Opportunities | Deal data sourced from the CRM |
search_conversations |
Conversations | ID, title, duration, creation date, account and person details |
conversation_by_id |
Conversations | Detailed conversation data including participants and messages |
conversation_recording_by_id |
Conversation media | Audio and video recording data for conversations |
search_users |
Users | List of users with ID and email |
user_by_id |
Users | User ID, GUID, name and email |
current_user |
Users | The Salesloft user interacting with the MCP client |
current_team |
Teams | The team the end user is associated with |
The conversation tools are the most valuable of the set. Being able to search conversations and then pull participants, messages and recording data is what makes the server worth connecting for a manager reviewing team activity.
Salesloft publishes sample prompts by persona, which is a useful way to see the intended shape. A rep asks for a summary of last week’s calls with an account and any objections raised. A manager asks which deals in a team’s pipeline are most at risk of slipping. A CRO asks about current pipeline revenue broken down by segment.
Every one of those prompts shares a property: it is a question. Nothing in this catalog takes an action, and that is by design rather than by omission.
Notice also what the object list does not include. There are no cadence, template, task or activity-creation tools, so the server describes accounts, people, deals and conversations rather than the outbound machinery around them. Teams expecting the conversational equivalent of their prospecting workflow will find this reads the results of that work instead of running it.
How to Set Up Salesloft MCP
Setup runs through an admin, then through each user’s AI client. The server stays dark until it is switched on for the organization.
Admin Configuration
An administrator switches the server on once for the whole organization, and the path runs through the AI settings rather than through an integrations menu.
- In Salesloft, open the Settings page as an admin and select the Artificial Intelligence settings under the Data section.
- Scroll to the Salesloft MCP section within those settings.
- Toggle the setting to enable the MCP Server, which activates the server discovery URL for your team. The toggle turns blue when it is on.
- Share the client-specific connection instructions with users, since the pre-configuration only makes connection possible rather than automatic.
Admin rights are required for all of this. Without them the toggle is not available, and neither is the discovery URL that clients need.
User Connection
Users authenticate with their existing Salesloft credentials, so there are no separate accounts to provision. Access is tied to the user’s Salesloft API key, and the supported clients per the help center are Claude Desktop and ChatGPT.
Permissions carry over intact from Salesloft itself. If you do not have access to a record in the platform, the assistant receives a 403 or an empty result when it tries to fetch that record, which means the server cannot become a way around your own permission model.
Revoking access has two levels. Removing or regenerating a user’s key in Settings under API Keys cuts off that individual, while turning off the organization-level toggle disconnects everyone at once.
If a connection fails, the documented cause of a 401 Unauthorized is an incorrect or revoked API key, and the fix is generating a new key and updating the client configuration.
Salesloft MCP Pricing and Availability
There is no public rate card here, and the MCP server sits outside the standard tiers as an add-on. The help center’s package availability grid marks it as unavailable on Advanced and Elite, and available as an Add-On.
The qualifying condition is narrower than that grid suggests. Salesloft states that the MCP server is available for Agentic customers only, meaning teams that have purchased one of the Agent packages such as Account Agent or Coaching Agent, who get the MCP server included.
There is a consumption dimension too. MCP calls route through the Salesloft API and count against your organization’s standard API rate limits, so heavy conversational use competes with your existing integrations for the same budget. That is worth modeling before rolling the server out to a full team.
The comparison a buyer actually needs is total cost against capability, and that is hard to run when one side publishes nothing. An Agent package plus the seats underneath it is a materially different commitment from a platform that lists its plans publicly, and the MCP server is rarely the deciding factor in either direction.