Skip to content
Back to skills

Iblai Api Agent Chat

ASecurity

Set up live chat with a deployed ibl.ai agent by wiring the hosted iblai-api-agent-chat MCP server into the project (.mcp.json / claude mcp add) using the .env Api-Token and a chosen agent's unique id. Use to enable runtime conversations with an agent. The actual chatting then happens through the MCP server's tools.

  • 15 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 20, 2026
ai-agentsbashgitapi

Works with

  • claude code
  • claude desktop
  • cursor
  • cli
  • api
  • mcp

Security analysis

A100/100

Scanned September 20, 2026

npx -y skills add iblai/vibe --skill iblai-api-agent-chat --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Iblai Api Agent Chat?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Iblai Api Agent Chat
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/iblai-iblai-api-agent-chat/badge)](https://www.skillsdirectory.com/skills/iblai-iblai-api-agent-chat)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: iblai-api-agent-chat
description: Set up live chat with a deployed ibl.ai agent by wiring the hosted iblai-api-agent-chat MCP server into the project (.mcp.json / claude mcp add) using the .env Api-Token and a chosen agent's unique id. Use to enable runtime conversations with an agent. The actual chatting then happens through the MCP server's tools.
metadata:
  kind: api
---

# iblai-api-agent-chat

Enable **live chat** with a deployed agent. Unlike the other `iblai-*` skills
(which call the REST API directly), runtime chat — streamed responses, tool use,
RAG — is served by the hosted **`iblai-api-agent-chat` MCP server**. This skill does
the one-time wiring so your assistant can talk to an agent; the conversation
itself then runs through that server's MCP tools.

## When to use

- You want to *converse with* an agent (not configure it). For configuring agents
  use the other `/iblai-api-agent-*` skills; to *find* an agent's `unique_id` use
  `/iblai-api-search`.

## What you need

- **`IBLAI_API_KEY`** — your Platform API Token (from `.env`; run `/iblai-api-login`
  if it is not set).
- **An agent's `unique_id`** — the agent to chat with. Get one from
  `/iblai-api-search` (each result's `unique_id`), then confirm the choice
  with the user.

## Setup

The server authenticates with two headers: `Authorization: Api-Token <key>` and
`X-Mentor-Unique-Id: <unique_id>` (one agent per connection).

### Claude Code

```bash
claude mcp add iblai-api-agent-chat \
  --transport http https://asgi.data.iblai.app/mcp/agent-chat/ \
  --header "Authorization: Api-Token $IBLAI_API_KEY" \
  --header "X-Mentor-Unique-Id: $MENTOR"
```

### .mcp.json / Claude Desktop / Cursor

Add to the project's `.mcp.json` (or the client's MCP config). Substitute the real
token and `unique_id` — do not commit secrets:

```json
{
  "mcpServers": {
    "iblai-agent-chat": {
      "transport": "streamable-http",
      "url": "https://asgi.data.iblai.app/mcp/agent-chat/",
      "headers": {
        "Authorization": "Api-Token YOUR_API_TOKEN",
        "X-Mentor-Unique-Id": "YOUR_MENTOR_UNIQUE_ID"
      }
    }
  }
}
```

After adding it, the assistant gains the server's chat tools — use those to send
messages and receive the agent's streamed replies.

## Notes

- **One agent per connection** — `X-Mentor-Unique-Id` is fixed per server entry.
  To chat with several agents, add multiple named entries (e.g.
  `iblai-api-agent-chat-tutor`, `iblai-api-agent-chat-advisor`), each with its own
  `unique_id`.
- The token lives in client config / `.env` — never commit it; `.env` and local
  MCP config should stay gitignored.
- This is the only runtime surface among the `iblai-api-*` skills; everything else (configure,
  search, recommend, profile, analytics, admin) is a direct-REST skill.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.