Lorikeet
by Lorikeet
Build, diagnose and test Lorikeet support agents from your AI assistant: workflows, guardrails, knowledge base, tickets and simulations. 106 tools, OAuth 2.1 sign-in, works with any MCP client.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Lorikeet via MCP
https://mcp.lorikeetcx.aiWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Lorikeet Tools & Capabilities (106)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- It manages your Lorikeet configuration, not your customers' conversations. The tools act on workflows, guardrails, integrations, articles and simulations. This is a connector for the team that builds the support agent, not one that answers tickets on a customer's behalf.
- A paid assistant plan is a real gate. Claude Pro, Team or Enterprise, or ChatGPT Team, Business or Enterprise, per Lorikeet's own prerequisites. ChatGPT additionally needs a Workspace Admin for setup.
- No per-tool documentation exists publicly. Lorikeet's MCP page covers setup, transport, auth and skills, but publishes no tool reference. We read 106 names from Anthropic's directory and no descriptions, parameters or safety annotations, because the endpoint requires OAuth.
load_promptis documented but absent from the directory listing. Lorikeet's skills page instructs clients to call aload_prompttool, and the directory's ownprompt_namesfield lists six prompts. That tool name is not among the 106 in the directory snapshot of 2026-08-16 — likely because MCP prompts are a separate protocol surface from tools. Expect the skill workflow to work as documented; do not expect to findload_promptin a tool list.- Unguided use degrades quietly. Lorikeet's documented failure mode for skipping skill invocation is "silent failures that look fine in the UI" — a caution the vendor publishes about its own connector, and the reason the skills section above is not optional reading.
- Most of the documentation site is private. Only the two MCP pages and the
llms.txtindex are publicly readable.docs.lorikeetcx.ai/and pages such as/tools/mcpredirect to a login, so broader product context could not be cited here. - We did not exercise any tool. Our check was an anonymous protocol request that returned an OAuth challenge. Everything above about tool behaviour comes from Lorikeet's documentation or from the directory listing, not from running anything.
Frequently asked questions
Which Claude or ChatGPT plan do you need for the Lorikeet MCP server?
A paid one. Lorikeet's setup docs list Claude Pro, Team or Enterprise for the Claude path, and a ChatGPT Team, Business or Enterprise subscription for the ChatGPT path. Claude Code has no plan requirement listed, only the CLI installed. Free-tier accounts on either assistant are not covered by the documented setup.
Do you need admin access to connect Lorikeet to your AI assistant?
For ChatGPT, yes. Lorikeet states Workspace Admin access is required for the initial setup, after which all workspace members can use the app. For Claude, its docs warn you may need admin access to add custom connectors and to contact your workspace admin if the option is missing. Claude Code needs neither.
Why does the Lorikeet MCP server produce worse results than Coach?
Because MCP has no automatic orchestration. Lorikeet's docs explain that Coach loads guided skills from your intent, while MCP exposes the underlying tools directly, so the assistant improvises. Its documented consequences are simulations created without test customer profiles, weak assertions, the wrong execution environment and silent failures. Invoke the skill explicitly instead.
How do you invoke a Lorikeet skill over MCP?
Explicitly, and the syntax varies by client. Lorikeet documents a slash command such as /lorikeet:create-simulations in Claude Code, and a universal fallback for Claude, ChatGPT, Cursor, Continue and Cline: ask the assistant to use the Lorikeet load_prompt tool to load a named skill, then describe the task.
How long does a Lorikeet MCP session stay authenticated?
One hour. Lorikeet's technical details table gives a token lifetime of 1 hour and states that interactive tokens auto-refresh while headless tokens are re-requested. Sign-in uses OAuth 2.1, with PKCE for interactive clients and client credentials for headless ones, so no API key is pasted into the assistant.
Do you need to pick a regional Lorikeet endpoint?
No. Lorikeet documents a single region-agnostic URL, https://mcp.lorikeetcx.ai, and states that authentication routes traffic to your account's region automatically. That holds regardless of where the account is hosted, so the same connect string works for every customer and no region-specific hostname is published.
Sources
- Live OAuth posture check against
https://mcp.lorikeetcx.ai/— an anonymousinitializereturns HTTP 401invalid_token, and the server publishes an RFC 9728.well-known/oauth-protected-resourcedocument (HTTP 200). This is our own observation on 2026-08-16, and it is why the sign-in requirement above is stated as verified rather than quoted from the vendor. It is also why no tool descriptions or parameter schemas appear on this page: the handshake never reachedtools/list. · retrieved 2026-08-16 - Lorikeet MCP Server documentation (HTTP 200, retrieved 2026-08-16 via the Markdown variant at
/mcp/mcp-server.md) · retrieved 2026-08-16 - Running simulations from MCP (HTTP 200, retrieved 2026-08-16 via
/mcp/skills.md) · retrieved 2026-08-16 - Lorikeet documentation index (HTTP 200, retrieved 2026-08-16; lists two public pages, both used above) · retrieved 2026-08-16
- Live auth posture check against
https://mcp.lorikeetcx.ai— HTTP 401 withWWW-Authenticate: Bearer realm="MCP"and an OAuth protected-resource pointer (2026-08-16) · retrieved 2026-08-16 - Anthropic Connectors Directory entry, including the 106 tool names, permissions and tier (from the committed directory snapshot of 2026-08-16) · retrieved 2026-08-16
- Lorikeet support — <support@lorikeetcx.ai> · Privacy (HTTP 200, checked 2026-08-16) · retrieved 2026-08-16
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Productivity
- Developer
- Lorikeet
- Tools
- 106
- Domain
- mcp.lorikeetcx.ai
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.