Chatley MCP Server: Manage AI Agents, Calls, Knowledge, and Campaigns
Managing an AI voice agent usually means logging into a dashboard, clicking through settings, and exporting call data by hand. The Chatley MCP Server lets you do that work from the AI client you already use. Connect Claude, ChatGPT, Cursor, or any other MCP-compatible client to your Chatley workspace, then ask it to review yesterday's calls, update an agent's configuration, or add a document to a knowledge base. This post covers what the server does, which tools it exposes, how you connect, and the safeguards that keep every change under your control.
What is the Chatley MCP Server?
MCP stands for Model Context Protocol, an open standard that lets AI clients use tools hosted on other systems. The Chatley MCP Server is a remote server that exposes your Chatley workspace as a set of tools an AI client can call on your behalf.
Every tool runs through the same internal service layer that powers the Chatley dashboard and Chatley Architect, the in-app AI configurator. An agent created through the MCP server therefore behaves the same way as an agent created in the dashboard, because both paths execute identical business logic, validation, and tenant scoping. The server adds the protocol layer, authentication, and tool annotations on top.
The launch set contains about 20 tools. Chatley kept the list short on purpose, since large tool lists slow down AI clients and reduce the quality of their responses.
Manage calls and analytics
Four tools cover call data. list_calls returns calls with filters and pagination, get_call_details returns the full record for one call including its outcome and metadata, get_call_transcript returns the transcript, and get_call_analytics returns aggregate results across calls.
This turns reporting into a conversation. You can ask your AI client to find every call from last week that ended without a booked appointment, read the transcripts, and summarize the 3 most common reasons. The client chains the tools together and returns a written answer instead of a spreadsheet you have to interpret yourself.
Analytics tools report value metrics such as calls handled, hours saved, resolution rate, and appointments booked. These are the numbers that show what your AI voice agent accomplishes for your business.
Manage AI agents
Six tools handle agent configuration. list_agents and get_agent read configuration, create_agent builds a new agent from a template, and list_voices returns the voices available to you. Voice IDs are case-sensitive, so the server returns them exactly as stored and your client can pass them back without alteration.
Editing follows a draft workflow. update_agent saves your changes as a draft and never edits the live agent directly. publish_agent_draft is a separate, explicit tool that pushes the draft live. This gives you a review point between "the AI client changed something" and "callers now hear the change." You can ask your client to rewrite an agent's greeting, read the draft back to you, and publish only after you approve it.
Manage knowledge bases
Knowledge bases hold the information your AI voice agent uses to answer callers, such as pricing, service areas, policies, and FAQs. Four tools manage them. list_knowledge_bases and get_knowledge_base show what exists and what each base contains, add_knowledge_base_item adds new content, and delete_knowledge_base_items removes it.
Deletion uses a two-phase confirmation. The first call previews the removal, and nothing is deleted until a second call carries confirmed=true. A single call can never remove data, so a misread instruction from an AI client cannot wipe out a knowledge base.
A practical use is keeping answers current. When your holiday hours change, you can paste the new schedule into your AI client and have it add the item to the right knowledge base in one step.
Manage contacts and campaigns.
list_contacts, create_contact, and update_contact manage your contact records. Creating a contact requires a consent state, so every record entering the system carries a record of whether you may reach that person.
list_campaigns shows your campaigns and their status, and trigger_campaign enrolls contacts into one. The server enforces consent and A2P messaging checks on every enrollment. If a contact lacks the required consent, the enrollment is rejected regardless of what the AI client requested. Compliance rules live on the server, where an AI client cannot talk its way around them.
How to connect your AI client
Connection uses OAuth with Dynamic Client Registration. Your AI client registers itself with Chatley automatically, so setup needs no manual client configuration, no provisioning ticket, and no sales call.
Open the connector or integration settings in your AI client.
Add mcp.chatley.ai/mcp as a remote MCP server.
Sign in to your Chatley account when the client prompts you.
Approve the access request.
Once connected, your client discovers the tools and their descriptions. Each tool carries annotations that mark it as a read, a write, or a destructive action. Clients use these annotations to approve safe reads automatically and ask for your consent only on writes, which keeps day-to-day use fast.
Built-in safeguards
An AI client that can change a live voice agent needs firm limits. The Chatley MCP Server enforces these on every call.
Draft before publish keeps agent changes off the phone line until you approve them.
Two-phase deletion requires an explicit confirmation before any data is removed.
Rate limits apply per connection and per tenant, so an AI client stuck in a retry loop cannot use up your workspace's capacity.
Full audit logging records the organization, tenant, tool, arguments, and result of every call, so you can always see what an agent did.
No secrets in responses means tool output never contains API keys, tokens, or third-party credentials, including inside error messages.
Delete tools for whole agents, contacts, and knowledge bases are excluded from the launch set, along with bulk updates by filter. Chatley will add them once there is demand and a confirmation flow proven in practice.
Manage multiple accounts from one connection
Agencies and partners often run dozens of client accounts. Every tool accepts an optional tenant_id parameter, so a partner administrator can act on any subaccount they manage through a single connection. list_tenants shows which subaccounts are in scope.
The server verifies authorization on every call. The server treats a tenant_id sent by the client as a request and checks that the authenticated organization administers that account before running anything. Cross-account access attempts are rejected and logged. An agency can operate a whole fleet of AI voice agents from one place while each client's data stays isolated.
Who should use the Chatley MCP Server?
Operations teams use it to pull analytics and adjust configuration without opening the dashboard. Business owners use it to manage their AI receptionist from the same AI assistant they use for email and planning. Agencies and partners use it to run many client accounts through one connection. Developers can use it to build repeatable workflows, such as a weekly report that reads transcripts and flags calls needing follow-up. If you prefer to manage agents from inside your Chatley account instead of an AI client, read our guide to AI agent management with Chatley Architect.