Claude / MCP connector

Lacudelph publishes a Model Context Protocol (MCP) server at https://mcp.lacudelph.com/mcp. Add it as a custom connector in Claude and your workspace becomes inspectable from inside a Claude conversation: list briefs, fetch transcripts, summarise rounds, publish briefs, create cohorts. Every tool call writes an audit row visible in /org/settings.

Setup

1. Go to /org/settings → Claude / MCP connector keys and create a key. Choose read-only for analysis (the safe default) or read + writeif you want Claude to publish briefs and create rounds. Copy the plaintext immediately — it’s shown once.
2. In Claude (web, desktop, or claude.ai/code), open Settings → Connectors → Add custom connector.
3. Paste https://mcp.lacudelph.com/mcpas the server URL. Choose “Bearer token” auth and paste your key.
4. Start a new conversation. Ask Claude to “list my Lacudelph briefs” — it’ll discover the tools and call lacudelph_list_briefs on its own.

Scopes

• read— list and fetch tools only. Cannot modify any state. Safe to leave attached indefinitely; safe to give to a Claude that’s helping non-admins explore the data.
• write — read tools pluspublishing briefs and creating rounds. Creating rounds is a Pro-tier feature and is bounded by your plan’s daily LLM cap. Write keys should be issued to a specific person and revoked when they’re done with the task.

You can have any number of keys. We recommend one per device or per Claude install so a leak only requires revoking that one key.

Tools

• lacudelph_list_briefs · read — List interview briefs in your org. Filter by status (draft / active / archived).
• lacudelph_get_brief · read — Fetch a brief by id, including its full template JSON (objectives, persona, takeaway shape).
• lacudelph_list_rounds · read — List cohort rounds. Filter by brief or status. Each row tells you whether an aggregate has been run.
• lacudelph_get_round_aggregate · read — Return a round's cross-cohort aggregate (themes, patterns, routing recommendations) plus its follow-up questions and conductor self-evaluation. Returns null fields if the aggregate hasn't run yet.
• lacudelph_list_interviews · read — List participant interviews. Filter by brief, round, or status. Excludes turn-by-turn transcripts — use get_interview_transcript for those.
• lacudelph_get_interview_transcript · read — Fetch a single interview by id with every turn (host + participant), the final takeaway markdown, and the latest extraction state. Use for deep analysis of a single conversation.
• lacudelph_set_brief_status · write — Publish, archive, or unpublish a brief. Publishing is gated by your plan's active-brief cap and refuses a brief with no template.
• lacudelph_create_round · write — Create an empty cohort round on a brief. Does NOT send invites or start interviews — only creates the shell. Pro-tier feature.

What the connector sees

Read tools return the same data the host sees in the Lacudelph UI: brief templates, round aggregates, completed interview transcripts. Participant transcripts are user-generated content — if you wouldn’t paste a participant transcript into Claude directly, don’t use the connector to fetch one. Anthropic’s data-use terms for your Claude plan apply to whatever the model receives via tool calls.

The connector does not send participant PII to telemetry. We log only tool name, status, and latency per call (visible in /org/settings) — never the inputs or outputs.

Error responses

• scope_insufficient — your key is read-only but the tool requires write scope. Issue a write-scope key and re-attach.
• rounds_require_pro — round creation is a Pro-tier feature. Upgrade from /org/billing.
• brief_not_found / round_not_found / interview_not_found — the id wasn’t found in your org. Cross-org access is impossible; the org is derived from your Bearer token, not from the tool input.
• cannot_publish_shell_brief — the brief has no template yet. Open it in the editor and run the design conversation first.
• tier_brief_limit_reached / free_tier_brief_limit_reached — your plan’s active-brief cap is full. Archive an old brief or upgrade.

Revocation

Revoke any key from /org/settings. Revocation is immediate — the next tool call on that key returns 401 from the auth layer before any tool handler runs.

Audit

Every successful and failed tool call records a row in the per-org audit log. brief.published, round.created, and similar state-changing events explicitly note source: mcp in their metadata, so a host reviewing audit history can tell which actions came from the connector vs. from the web UI.

Limitations (v1)

• The connector does not start new interviews or send invite emails. create_round creates the round shell; you still distribute the participant URL yourself (or via the existing brief share UX).
• The connector does not author new brief templates. Use the design-conversation in the Lacudelph UI to create the brief; Claude can then publish it via set_brief_status.
• No streaming. Tool calls return when the query completes; nothing is incremental.