Motif MCP server

Motif

<!-- mcp-name: io.github.sleepycobalt/motif -->

An agentic loop that turns a folder of interview transcripts into a research synthesis where every insight carries cited, verified evidence, an honest confidence level, and the counter-evidence against it.

Built for design and research teams who synthesise qualitative interviews and need output they can trust and trace. Motif is the first tool from ETOT. Built as an R&D project; the case study tells the story.

What it does

transcripts/  →  intake  →  synthesis  →  critic  →  revise  →  report.md
                                            ↑            │
                                            └────────────┘  until the critic passes or 3 rounds

• Intake (one call per transcript) maps topics and notable positions with turn references.
• Synthesis produces 8–14 insights. Each has a claim, cited turns with verbatim receipts, sources, confidence, counter-evidence, and a design opportunity.
• Critic checks every insight against the transcripts using rules you can edit — unsupported claims, missing dissent, overconfidence, merged findings, themes present in the corpus but absent from the report. Some rules run in code (citations exist, quotes match, confidence thresholds); the rest are judged by the model.
• Revise fixes what the critic flagged. It may not delete an insight to make an objection go away.
• Report shows every insight with its evidence expanded, and marks any insight the critic still objected to when the loop stopped. Silence is never treated as agreement.

Sample output: docs/exhibits/best-report-v2/output.md.

Install (about 5 minutes)

You need Python 3.10+ and an Anthropic API key (console.anthropic.com).

pip install etot-motif
export ANTHROPIC_API_KEY=your-key-here      # or put it in a .env file in the working directory

Or from a checkout, if you want to edit the critic rules or run the evals:

git clone https://github.com/sleepycobalt/motif.git
cd motif
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
echo "ANTHROPIC_API_KEY=your-key-here" > .env

Run

Put your transcripts in a folder, one speaker turn per paragraph or line, each starting with the speaker's name and a colon. Label the interviewer Researcher, Interviewer, or Moderator so their turns are never cited as evidence.

Interviewer: Can you tell me about the last time you used the app?
Priya: Sure. I opened it on the train and it logged me out again, which...

Then:

motif ./transcripts --out report.md --question "What frustrates users about onboarding?"

Fifteen transcripts of ~45 minutes each take about 20 minutes and cost about $2.50 in API usage. Every prompt, response, and iteration is saved under runs/ so you can see exactly what the critic objected to and how the synthesis changed.

Try the sample corpus first:

motif data/raw/Dataset-2 --out report.md

Use it from Claude Code or Cursor

Motif is also an MCP server: the same engine, callable from any MCP host.

pip install "etot-motif[mcp]"
claude mcp add motif -e ANTHROPIC_API_KEY=your-key-here -- motif-mcp

Or, with uv and no install step, claude mcp add motif -e ANTHROPICAPIKEY=your-key-here -- uvx --from "etot-motif[mcp]" motif-mcp. From a checkout: pip install -e ".[mcp]" and point the host at $PWD/.venv/bin/motif-mcp.

Shortened. The full README is on GitHub.