Claude Code

The Claude Code CLI supports MCP natively. Once AssureAgent is added, claude can send messages to Vibe (your QA developer) and read back replies inline.

Prerequisites

Claude Code installed (npm i -g @anthropic-ai/claude-code) and signed in.
A personal AssureAgent API key. Mint one from the portal: Settings → API Keys → Create key. It starts with aa_live_.

Add AssureAgent as an MCP server

In your terminal, run:

claude mcp add assureagent --transport http https://mcp.assureagent.ai/mcp \
  --header "Authorization: Bearer aa_live_YOUR_KEY"

Restart the claude session (or open a new one). List servers to confirm:
Terminal window
```
claude mcp list
```
You should see assureagent listed.
Ask Claude to use it:
```
> Ask my QA developer what projects I have.
```
Claude will call list_projects on the AssureAgent MCP server and summarise the response.

What Claude can do through MCP

The AssureAgent MCP server ships with two agent-to-agent tools and a handful of read shortcuts:

talk_to_vibe

The primary tool. Claude sends Vibe a natural-language message; Vibe replies in prose. Use this for anything that requires judgment — scenario design, running tests, interpreting results.

Reuse the returned conversation_id for multi-turn work; omit it to start a fresh conversation.

Example prompts to try:

“Ask my QA developer to design a scenario where Sarah calls +1 980-414-3834 about a personal loan and gets transferred to a specialist.”
“Have Vibe run the last scenario in project Trexis FNOL and share the report link when it’s done.”
“Ask my QA developer which runs failed this week and why.”

poll_conversation

Companion to talk_to_vibe. When Vibe returns status: 'working' (long-running work like a test call), Claude calls poll_conversation after ~10s to check for the reply.

list_projects

Lists your AssureAgent workspaces. Useful when Claude needs a project_id to disambiguate.

list_recent_runs

Recent voice + chat runs interleaved, with pass/fail verdicts. Great for stand-up recaps (“what did we run yesterday?“).

list_scenarios

Scenarios in a project (or all), with target, type, and last-updated info. Feed the name to talk_to_vibe to run one.

Where MCP conversations appear in the portal

Every conversation Claude has with Vibe lands under Manager → AI conversations in the portal, badged Claude Code. Same bucket counts against your monthly turn cap.

Troubleshooting

invalid or revoked api key — the key you pasted is malformed or was rotated. Mint a fresh one and re-run claude mcp add.

Claude doesn’t call talk_to_vibe when I ask — try a more explicit prompt: “Use my AssureAgent QA developer to …”. Once Claude sees a few examples in the conversation, it settles into the pattern.

Long-running work timing out — if Claude gets a status: 'working' reply and then stops, ask it to call poll_conversation again with the returned conversation_id.