LLM + MCP Guide

Start using Scribe MCP in minutes.

Scribe MCP exposes one thin CLI bridge: pass a command and its arguments, then use the structured result.

Open full LLM.md About Scribe

Get Started

First 5 minutes

1. Connect to Scribe MCP

Point your MCP client to Scribe and verify tool discovery.

scribe_command({ command: "help", args: [] })

2. Find relevant transcriptions

Search before loading large records.

scribe_command({ command: "search", args: ["roadmap", "--limit", "10"] })

3. Load minimal context first

Use summary and metadata before segments.

scribe_command({ command: "summary", args: ["<id>"] })

Golden rule

Search first. Summary second. Segments last.

This keeps token usage low and response quality high.

Recommended request pattern

1) scribe_command({ command: "search", args: [query, "--limit", "10"] })
2) scribe_command({ command: "summary", args: [id] })
3) if needed: scribe_command({ command: "transcript", args: [id] })
4) stop when the question is answered

Typical segment page: 100-200

Avoid loading: 1000+ at once

CLI command map

Transcriptions

8 commands

Search, load, segment, summarize

Documents

4 commands

Create, link, export, collaborate

People & Companies

9 commands

Directory and entity context

Good prompt to start with

"Find the latest meeting where roadmap priorities were discussed, summarize decisions, and list explicit owners and deadlines. Use summaries first, then only load segments if needed."

Quality checklist

Always cite which transcription(s) were used
Distinguish explicit decisions from inferred suggestions
Stop loading data once the user question is answered

Need the full reference?

The complete technical guide, examples, and edge cases are available in the repository docs.

Open LLM.md on GitHub