Publish one assistant surface
Admins define which objects, fields, bundles, and tools one assistant is allowed to use.
Salesforce MCP server for teams that need boundaries.
emcee is a remote Salesforce MCP server and control layer for bounded assistants. It uses Model Context Protocol to expose assistant-specific Salesforce tools instead of giving a generic AI client broad org access.
The current runtime endpoint is https://runtime.emcee-ai.com/mcp
for approved early-access customers. Transport is Streamable HTTP,
and the runtime surface is intentionally narrower than a generic
Salesforce connector.
No raw SOQL passthrough. One assistant per team. Named-user runtime binding.
What emcee is
- A remote Salesforce MCP server on
/mcp. - A bounded tool surface over Salesforce REST.
- A publish-before-runtime model for assistant-specific access.
- A named-user runtime with tenant, org, user, and assistant binding.
- An early-access product, not a generic whole-org connector.
Recruitment assistant setup
Real product screenshot from the current assistant setup flow.
Why this is not a generic Salesforce connector
Many Salesforce MCP servers focus on broad object access, record search, or free-form query power. emcee is built around a narrower problem: giving one team a useful assistant without opening the whole Salesforce org to an AI client.
Assign named users
Assistant access is granted deliberately to people, not assumed from a shared connector.
Bind runtime per assistant
Each runtime session resolves to one tenant, one org, one user, and one assistant.
Log runtime events
Invocations emit structured audit events instead of disappearing behind a generic chat session.
Current Salesforce MCP posture
The current remote runtime is intentionally specific. It supports bounded retrieval, document and note operations, and assistant-scoped context assembly. It does not expose generic SOQL passthrough.
Transport
Streamable HTTP on /mcp behind the public runtime host https://runtime.emcee-ai.com.
Runtime model
Remote MCP runtime for approved Claude and OpenAI-style client connections rather than a local desktop-only setup.
Tool surface
Schema discovery, structured listing, structured aggregates, record bundle assembly, and bounded note or document actions.
What it does not expose
No raw SOQL passthrough, no generic whole-org search surface, and no pretending the endpoint is broader than it is.
Salesforce MCP FAQ
These are the plain questions people ask when they are comparing MCP servers, Salesforce connectors, or early-access runtime products.
Is emcee a Salesforce MCP server?
Yes. emcee exposes a remote MCP runtime for Salesforce through Model Context Protocol, but the surface is bounded per assistant instead of broad by default.
What is the MCP endpoint?
The current runtime host is https://runtime.emcee-ai.com and the MCP server URL is https://runtime.emcee-ai.com/mcp for approved early-access customers.
Does it expose raw SOQL?
No. The current runtime uses Salesforce REST underneath and intentionally avoids generic SOQL passthrough as a product surface.
How is access scoped?
Admins publish one assistant surface, assign named users, and runtime sessions bind to one tenant, one org, one user, and one assistant.
Which teams are the best fit first?
Recruitment is the clearest live example today. Sales and service follow the same pattern when the problem is safe retrieval, not a general chatbot.
Is this self-serve already?
No. emcee is early access and the rollout is still hands-on. That is deliberate and part of the current product posture.
When this page is useful
- You are searching for a Salesforce MCP server, not a generic AI demo.
- You need a remote runtime endpoint instead of a local-only connector.
- You need boundaries, named-user access, and a security answer that stands up in review.