Connect Guzli to Claude and ChatGPT via MCP
Guzli has a remote MCP server at https://mcp.guzli.com/mcp. In a host that supports remote MCP connections, you can connect the server, complete Guzli sign-in, and work with the tools exposed to your account. The connection lets you inspect and manage Guzli work, while checking what a tool will change and whether approval is pending.
Quick verdict
Add Guzli’s remote MCP server in your host’s connector settings and complete OAuth sign-in. List the exposed tools and schemas before making changes. Keep sending, spending, number releases, and system-prompt changes behind authorization and approval.
Core tools cover contacts, tags, segments, knowledge ingestion and status, managed phone numbers, agent configuration, conversations, and issue reports. Email outreach and voice campaigns have their own tool families.
Who this is for
Use the connection when you already work in Guzli and can name the task and its scope.
- Team members finding a contact, reviewing tags, or previewing a segment before outreach.
- Operators checking whether an agent’s knowledge source finished ingesting.
- Campaign owners reviewing a draft and its intended audience before requesting a send.
The returned identifiers and outcomes help you verify each next step. A conversational request can organize the work, but those facts still matter.
Who this is not for
The connection should not be used to skip permissions or approvals.
- A sender without a verified channel and purpose permission for the intended contacts.
- Someone treating an accepted ingestion task as completed knowledge or a queued campaign as delivered.
- Someone treating a held tool call as an approved change or retrying an uncertain action.
Connector controls vary by host. The shared setup point is the Guzli server URL and OAuth sign-in.
The playbook: connect in five steps
Add the remote server
Open your host’s connector settings and add a remote MCP server using
https://mcp.guzli.com/mcp. Use the host’s own connection flow; no host-specific menu path is needed here. The Guzli server is namedguzliin plugin configurations, and fully qualified tools use theguzli:<tool>form.Complete Guzli sign-in
Finish the Guzli OAuth sign-in in the host. After the connection is available, inspect the exposed tools and their argument schemas. If a tool you expect is absent, reconnect and list the tools again before acting. The visible schema, rather than an example copied from an article, defines the arguments for that session.
Start with a read
Ask for a specific read, such as finding a contact or inspecting an agent’s tool configuration. For a contact lookup, distinguish a match returned by Guzli from a name or address mentioned in conversation. For a segment, inspect the field catalog and preview the audience before creating or using it. Keep the returned contact, agent, segment, and version identifiers distinct.
Prepare one bounded change
Name the object, the requested change, and its scope. For knowledge, provide the agent and an authorized URL, submit the ingestion, then keep the returned task ID. Check ingestion status for that task; acceptance means processing has started, not that the source is ready. For outreach, review the draft, intended audience, permission basis, and sender before any publish or run step.
Inspect the outcome
After a change, read back the affected state when a read tool exists. If a tool returns a hold, report that approval is pending and use the host’s approval flow. Do not repeat an uncertain send, call, purchase, release, or submission simply to get a success response. A completed operation identifier and an approval hold identifier describe different states.
Use cases
Enrich a contact list from a conversation
Begin by finding existing contacts and resolving ambiguous matches. Add only facts the user supplied; do not invent an email address, phone number, or permission basis. You can inspect tag definitions, create or edit a requested tag, and attach it to the right contacts. If the list should become a reusable segment, inspect allowed fields, preview matches, create the authorized expression, and check readiness before using it in outreach.
Ingest a URL into the knowledge base and check status
Give Guzli the intended agent and a specific URL. The core tools can submit a single-URL ingestion and return a task ID. Check that task’s status instead of assuming the content is already available. When ingestion completes, search the knowledge base for a source-specific question and confirm the returned evidence supports the answer. If ingestion fails, inspect the error and current state before submitting another task.
Draft an outreach sequence that waits for approval
The email outreach tools can create and edit a campaign draft. Define the audience, purpose, sender, content, postal address, and daily cap, then inspect the saved revision. Stop at the draft if that is all the user requested. Publishing, enrolling, running, and sending are separate steps that need authorization and permission checks. Voice campaigns follow a similar draft and readiness flow, with call instructions, caller selection, and result inspection.
FAQ
Which address do I use?
Use https://mcp.guzli.com/mcp as the remote MCP server address. Complete Guzli OAuth sign-in through the host after adding it.
Can the connection send email or place calls?
Guzli has sibling tool families for email outreach and voice campaigns. They can prepare campaigns and, when authorized and ready, send or call. The host should inspect the exposed tool schemas and check recipient permission and the actual result before reporting an action as complete.
Can it buy or release a phone number?
Core tools can search, buy, list, and release managed numbers. A purchase requires an identified selection and explicit spending authorization. A release requires the exact owned number and authorization, and it follows an ask-first approval flow. Verify inventory after either action.
Can it change an agent’s instructions?
The tools can inspect agent configuration and configure a system prompt. Replacing a system prompt needs the full intended text, the intended agent, and the engine’s ask-first approval. A held request is pending, not a saved configuration.
How do I know whether a task finished?
Read the operation’s returned status and then use the relevant read tool. Knowledge ingestion has a task ID and status check. Campaigns expose draft, readiness, enrollment, and measurement information. For an uncertain mutation, inspect the affected object before considering a retry.