Build agents your way: Introducing Parloa’s Agent Management API, SDK, CLI, and MCP

Your teams have invested years in the tools and workflows they trust. Why should building AI agents mean leaving that investment behind? 

New tools work best when they connect to the systems teams already run. If AI agent management sits outside those workflows, teams can end up redoing much of the work they have already done. As a result, deployment and optimization gets stalled by copying settings, repeating configuration, and maintaining custom workarounds. 

When engineers can manage agent operations through their existing code and release pipelines, and agent builders working with AI assistants can develop agent changes alongside their application and integrations, however, teams spend less time starting over and more time scaling. That’s why Parloa is adding more ways to build alongside its no-code AI Agent Management Platform with Agent Management API , so teams can manage agents from the tools they already use.

For engineers and technical integrators, the new Agent Management API provides access to build, configure, test, and deploy agents. They can call it directly, use the Python SDK in their code and release pipelines, or use the command-line interface (CLI) for ready-made terminal commands. For agent builders, Parloa MCP connects supported assistants such as Claude Code, Cursor, and ChatGPT to the same agent-building capabilities. Builders can review, change, and test agents by describing what they want in plain language through the assistant they already work in.

Consider a service-policy change. Customer experience (CX) defines the intended behavior and acceptance criteria. Engineers get the agents up and running through the API, SDK, or CLI. They update the affected agents with a script, run simulations and evaluations in the release pipeline, and deploy the version that passes. Builders optimize and scale from there. They request the next change through a connected AI assistant and review the preview before it is applied. Different access routes, same agents, and the same review, testing, and release steps your team defines before anything touches a customer.

Bring agent operations into your engineering workflows

For engineers and integrators who own testing, reliability, and releases, the Agent Management API, SDK, and CLI make agent management part of the engineering process they already run. With the API, SDK, and CLI, engineers are able to:

• Automate repeatable work. Create, update, and delete agents, skills, knowledge, deployments, and channels through the API, Python SDK, or CLI. Turn recurring operations into reusable scripts, and define a change once to apply it across affected agents.

• Keep versions where everything else is tracked. Store agent settings in the version control they already use, publish each change as a version, and promote it from testing to live, with rollback as needed.

• Release only what passes. Run simulations and evaluations from their release pipeline. Check the results against the acceptance criteria your business defines, and configure the pipeline to block releases that fail. Teams that already use Parloa’s automated testing can turn it into a required step in every release, so no agent reaches customers without passing the test.

Update agents from the AI coding agents your builders already use

Parloa Navigator already lets teams build and optimize AI agents using natural language and without code. But until now, builders using other AI assistants like Claude or ChatGPT for the rest of their work couldn’t use the same assistant to update a Parloa agent. Parloa MCP closes that gap.

Now, builders can describe what they want in plain language, and their preferred AI assistant will help them review the agents, change prompts, test, compare versions, and prepare deployments. Each change is previewed before it’s made and saved as a draft. A coding assistant can combine agent work with the documents, SOPs, tickets, and other MCPs the builder has already connected. Builders can use that context to request agent changes without repeatedly copying prompts and configuration between tools.  

Control stays part of every workflow

Managing agents from code and AI assistants is only useful if teams can trust how an automated change reaches production. That’s why we built control into both routes:

• Agent builders sign in as themselves. Parloa MCP uses SSO, so the assistant can only do what that builder is already permitted to do in Parloa’s platform. There’s no API key to create, store, or revoke.

• Pipelines get scoped access. An Admin creates the API client, with limited access and short-lived tokens, scoped to one organization and one region.

• Read-only by default. MCP access starts as read-only, and write access must be turned on explicitly.

• Edits made in the assistant are previewed and saved to a draft. Agent configuration changes made through MCP are previewed before they’re applied. 

• Publishing and deployment remain separate actions. Teams can configure  review and testing steps in their release workflows.

One agent program with more ways to manage it

Introducing AI agents for your CX workflows shouldn’t mean leaving the tools and workflows your teams have spent years building and working with behind.

When agent management fits the workflows teams already run, an AI program can reliably grow from a few agents to a portfolio without rework, with more people contributing, and more time to improve customer service.

Interested in learning more about Parloa’s Agent Management API, SDK, CLI and MCP? Talk to our team.