--- title: "Talk to Any ibl.ai Agent From Claude Code, Cursor or Claude Desktop" slug: "builtin-mcp-servers-agent-chat" date: "2026-11-05" tag: "Application" summary: "Every agent you deploy on ibl.ai can now be added to Claude Code, Cursor, Claude Desktop or any MCP client in one step, and talked to from there: the agent answers with its own knowledge, memory and guardrails. Here is the two-minute setup and a real conversation with a university's advising agent." author: "ibl.ai Engineering" linkedin: | Every agent on ibl.ai can now sit inside the tools people already use. Add an ibl.ai agent to Claude Code, Cursor or Claude Desktop in one step, and talk to it from there. It is the same agent you built in Agentic OS, with the same knowledge, memory and guardrails. Not a generic chatbot. Your admissions agent, your advising agent, your policy agent, reachable from wherever your people work. We tested it with a university's advising agent today. A student named Dana asked how to transfer, how many of her 45 credits would count, and then asked the agent to recap. It remembered her name, her question, and her credits. The setup is one agent ID and one API key. No integration project. And on ibl.ai you own all the code and the data, so the agent your team talks to from Claude Code is one that lives inside your own perimeter. #iblai #AgenticAI #EnterpriseAI #MCP #HigherEducation --- **Every agent you deploy on ibl.ai can now be added to Claude Code, Cursor, Claude Desktop or any other MCP client in one step, and talked to from there, with its own knowledge, memory and guardrails intact, because on ibl.ai you own all the code and the data, and now your agents go wherever your people work.** MCP, the Model Context Protocol, is the plug that lets AI tools talk to other software. Claude Code, Cursor and Claude Desktop all speak it. As of this release, so does every agent on [Agentic OS](/product/agentic-os). You add the agent once, and from then on you can ask it questions without opening another window. ## Which agents can I add? Any agent you have deployed on ibl.ai. Each one is added on its own, so what shows up in Claude Code or Cursor is the specific agent you chose, exactly as you configured it: its instructions, the documents it is grounded in, its memory and its guardrails. - A university's **admissions, advising or financial-aid agent** ([higher education](/solutions/higher-education)) - A hospital's **clinical-support or prior-authorization agent** ([healthcare](/solutions/medical-healthcare)) - An agency's **case-work or compliance agent** ([government](/solutions/government)) - Any agent you built from your own documents Add more than one, and the tool you are working in can ask the right agent for each question. The advising agent answers the student questions; the policy agent answers the policy questions. ## How do I add an agent? You need two things, both from Agentic OS: the agent's ID (it is in the agent's settings) and an API key for your organization (created from your account settings). Then it is one step. In Claude Code, one command: ```bash claude mcp add --transport http my-advising-agent \ https://asgi.data.iblai.app/mcp/agent-chat/ \ --header "Authorization: Api-Token " \ --header "X-Mentor-Unique-Id: " ``` In Cursor or Claude Desktop, paste this into the app's MCP settings: ```json { "mcpServers": { "my-advising-agent": { "url": "https://asgi.data.iblai.app/mcp/agent-chat/", "headers": { "Authorization": "Api-Token ", "X-Mentor-Unique-Id": "" } } } } ``` That is the whole setup. From then on you ask in plain language: "Ask my advising agent what a transfer student needs." Give the connection a name that says which agent it is, because you can add several. The key belongs to your organization and the agent has to belong to the same one, so nobody can reach an agent from outside the organization that owns it. On a self-hosted deployment the address is your own server, not ours. ## What does a conversation look like? We added a university's advising agent, Campus Advisor, and had a short conversation with it while writing this post. Nothing below is edited.
Dana: Hi, I'm Dana. I'm a community college student thinking about transferring to Lakeview next fall. What should I do first? Keep it to three sentences.
Campus Advisor: Hi Dana! First, contact Lakeview's transfer admissions office to confirm application deadlines, required materials, and whether they have a transfer guide for your community college. Next, meet with your community college transfer advisor and request an official credit evaluation so you know which courses will transfer. Also check Lakeview's financial aid and scholarship deadlines early so you don't miss funding opportunities.
Dana: I'll have about 45 credits by then. Will most of them transfer, and who decides? Three sentences max.
Campus Advisor: It depends on how your courses match Lakeview's requirements, your grades, and any limits on transfer credits, so 45 credits may not all apply toward your degree. Lakeview makes the official decision, often through its transfer-credit or registrar's office and your academic department. Ask for a preliminary credit review and share your transcript and course syllabi.
Dana: Thanks. Can you remind me what my name is and what I asked you about, in one sentence?
Campus Advisor: Your name is Dana, and you asked about transferring to Lakeview next fall, including how to get started and whether your roughly 45 credits will transfer.
Three things to notice. The agent answers as the university's advisor, not as a general chatbot. It follows the conversation: by the third turn it knows Dana's name, her plan and her credits. And every answer came back through the same guardrails the agent has in Agentic OS, because it is the same agent. ## Why does this matter? Most organizations have the same problem: the agent is good, but people have to go to it. Now the agent comes to them. A registrar working in Cursor, an analyst in Claude Desktop, a developer in Claude Code, all get the institution's own agents without leaving their tool, and without anyone building an integration. The agent you expose this way is the one you govern. Its knowledge, its memory, its limits and its audit trail stay yours, inside your organization. On ibl.ai you own all the code and the data. For the full technical reference, including the other platform tools now available over MCP, see [MCP Servers](/developer/platform/mcp/mcp-servers) in the developer docs.