ChatGPT MCP Capabilities
J
Josh Carroll
I was able to connect this MCP, but after review I found the connected Vendasta plugin currently exposes:
- CRM contacts, companies, opportunities, associations, fields, and pipelines
- Both read and write CRM operations
- Vendasta documentation search
It does not currently expose tools for directly reading or updating:
- AI employee role prompts
- Capability prompts
- Knowledge sources
- Channels
- Autonomy levels
- AI Workforce activation settings
I really need it to allow access to those additional tools. Can we do that?
Log In
H
Harshita Garge
Hi Josh,
Thanks for testing the MCP and for the detailed breakdown. Your read is correct. Today the Vendasta MCP covers CRM (contacts, companies, opportunities, associations, fields, pipelines) with read and write access, plus documentation search. AI Workforce configuration isn't exposed yet.
we plan to expose more tools and allow more usecases via MCP in upcoming quarter.
I can't commit to a timeline right now, but I want to make sure we build the right thing if we add this. A few questions:
- What are you trying to get done? For example, setting up AI employees across many client accounts at once, keeping prompts in sync from an outside source, auditing current configs, or something else.
- Do you need to read these settings, change them, or both? Reading prompts and knowledge sources is a very different risk from letting an AI change autonomy levels or turn on an AI employee.
- Roughly how many accounts or AI employees would you manage this way?
What are you doing today instead? If it's manual work in the UI, how long does it take you?
Your answers will help us decide what to expose first and what guardrails it needs.
I've logged this as a request for AI Workforce tools in the MCP.
Thanks,
Harshita
J
Josh Carroll
Harshita Garge,
We’re building an internal Vendasta AI Workforce Deployment system for SnapMe Creative Agency in ChatGPT. The goal is to give our employees one AI agent that coordinates the onboarding, configuration, testing, launch, and ongoing review of Vendasta AI Employees across our client accounts.
We’ve built specialized deployment workflows for:
Chat Receptionist
Voice Receptionist
Social Media Manager
Blogger
Reputation Specialist
Search Specialist
Sales Assistant
These are not standardized configurations pushed across clients. Each AI Employee is built specifically for that client’s business, services, brand, objectives, processes, escalation requirements, and approved scope.
Our intended workflow is:
Approved Scope → Client Discovery → AI Employee Configuration → Human Review → Vendasta Configuration → Read-Back Verification → Independent QA → Human Launch Approval → Activation → 7/14/30-Day Reviews
Ideally, we need both read and write access.
For read access, we’d like to retrieve each AI Employee’s:
Role and capability prompts
Knowledge sources
Channels/integrations
Escalation settings
Autonomy settings
Activation/status information
Read access alone would be extremely valuable because our QA process could independently audit what is actually configured in Vendasta against what SnapMe approved.
J
Josh Carroll
Harshita Garge,
For write access, we’d like to update those configurations after human approval. We would actually prefer guardrails around higher-risk actions. Updating prompts or knowledge on an inactive employee could be treated differently from changing autonomy, modifying a live employee, or activating one.
Our ideal write process is:
Prepare Change → Human Approval → Execute → Read Back → Verify
We also want every action tied to a specific Vendasta Business/Account ID to prevent accidental cross-client changes.
Today, ChatGPT can help us gather requirements, build prompts, determine knowledge requirements, establish escalation rules, and create testing plans. But a SnapMe employee must then manually transfer everything into the client’s Business App, verify it, test it, and activate it. That manual handoff takes time and introduces the possibility of configuration errors.
Initially, we expect dozens of client accounts, but a single client may have several AI Employees, so this could quickly become 100–300+ individual AI Employee configurations under management.
If you’re prioritizing what to expose first, read access to the complete AI Workforce configuration would immediately help us. Controlled write access to inactive/draft configurations would be our next priority, followed by autonomy and activation controls with explicit human approval.
We’d also be happy to serve as a real-world testing partner as these MCP capabilities are developed.
Josh