Docs

What Ciev.ai is, how to connect it, and what its tools do.

What it is

Ciev.ai ingests selected engineering meeting recordings — architecture reviews, onboarding walkthroughs, design discussions — and turns them into searchable, timestamped evidence: transcript and on-screen content alike. It's exposed to your coding agent as a set of MCP tools, plus one tool that renders screenshots and clips inline for hosts that support it (like Claude).

Connect your agent

Claude, or any MCP client with OAuth support: add https://ciev.ai/mcp as a remote MCP server. You'll be prompted to sign in to your Ciev.ai account (or connect with Google) and approve the connection — no API key needed.

Claude Code, Claude Desktop, or a client without OAuth support: use an API key instead, from your Ciev.ai admin area under Sources → API keys.

{
  "mcpServers": {
    "ciev": {
      "command": "npx",
      "args": ["-y", "ciev"],
      "env": {
        "CIEV_API_KEY": "",
        "CIEV_BASE_URL": "https://ciev.ai"
      }
    }
  }
}

Tools

search_recordings

Semantic search over meeting chunks. Compact, cited results (title, date, timestamps, summary) — safe to call speculatively.

grep_context

Exact substring match over transcript and on-screen text alike, for identifiers and exact phrases.

read_context

Reads the full annotated evidence for one or more chunk IDs, optionally including the neighboring chunk on each side.

list_meetings

Lists real recordings in a workspace (meeting ID, title, date, short summary) — resolves a recording by name instead of requiring a search hit first.

get_document

Fetches a meeting's synthesized document — its Digest (decisions, open questions, action items) or, for a coding-interview recording, its interview report — instead of raw transcript.

get_frame_images

Fetches the actual pixels for specific frames, for cases where a caption or on-screen text alone isn't enough.

get_clip

Returns a plain link into the original recording, seeked to a chunk's timestamp span.

render_visual_answer

Renders an answer inline with real screenshots and clips embedded — the one tool that shows evidence directly to you, on a host that supports it.

Every response carries the source meeting's title, date, and timestamps — an answer without that citation is a bug, not a formatting choice.