MCP

Ready for your agentic workflow.

Observant installs as an MCP server, right inside Claude Code or Cursor — built for AI-native teams who ship in the editor. User learning runs on autopilot and shows up where you build: surfaced in the flow, agent-ready, already half-implemented.

claude code — ~/northwind observant · mcp
why are power users dropping the new export flow? observant  pulling from your panel… triggered by PostHog · 22% drop-off on export_v2 · cohort: Power users 18 recent 1:1s matched · synthesizing
Insight · agent-ready
61% rebuild the export by hand to get a shareable view. The CSV-first flow is the friction — not the export itself.
→ recommend: ship a live, shareable dashboard; keep CSV as secondary
3 supporting quotes · draft spec + Linear issue attached
draft the implementation plan

No dashboard, no context-switch — the insight arrives where you're already building.

Closed loop

From a signal to a shipped fix — without leaving your editor.

Detect
PostHog spots it
A drop-off or behavior change fires the trigger.
Learn
Observant runs the 1:1s
Real conversations with the exact users affected.
Synthesize
Proactively, not on ask
Observant surfaces the pattern before you think to query it.
Implement
Your agent ships it
Agent-ready insight → spec → code, in the same session.
One door

Your users' feedback and the research engine — in the same MCP.

Connect once. Observant gives your editor agent two things: the raw feedback your users have already given you, and the engine that runs new studies and synthesizes it. No dashboard to open, no export to wrangle.

Observant data · live

Pull your feedback

What your users actually said — read it where you build.

  • observant_recent_feedbackIn-product snippet feedback — thumbs, exit reasons, short notes, CSAT.
  • observant_conversationsOff-product 1:1 threads (email/Telegram) with every turn, per named user.
  • observant_insightsSynthesized Signals — Issues, Insights, Opportunities, grounded in quotes.
Codified engine · bridged

Run the research

Orchestrate new studies over that same feedback, from the editor.

  • codified_run_studyKick off a round — churn, drop-off, PMF, launch feedback. Observant recruits + runs the 1:1s.
  • codified_query_insightsSearch synthesized insights by feature or topic across studies.
  • codified_search_transcriptsFull-text search every interview transcript for what users said.
Connect

One command to wire it into your editor.

Add the MCP server over HTTP — Claude Code, Cursor, or Claude Desktop. No dashboard, no setup call.

$ claude mcp add --transport http observant https://www.observanthq.com/api/mcp

Cursor & Claude Desktop: point an HTTP MCP server at the same URL. Want studies too? Add the engine alongside it:

$ claude mcp add codified -- npx -y @usercodified/mcp
Read the docs
Plugs into

Where your team already works.

CC Claude Code Cu Cursor CD Claude Desktop Ph PostHog Li Linear No Notion Sl Slack { } Webhooks