# FeedbackGraph

FeedbackGraph is an AI-native bug and feedback platform for B2B SaaS teams. It captures bug reports and customer feedback in two clicks (screenshot, screen recording, console/network/error context), enriches each report with AI that shows its evidence beside the verdict, merges near-duplicate reports, and routes the work to Jira, Linear, GitHub, and Slack with two-way status sync — then closes the loop back to the original reporter. Teams use it to catch user-reported bugs before they reach customers, speed up QA and UAT, ship more stable releases, and grow revenue by building what accounts actually pay for. A revenue-backed Evidence Graph rolls every signal up to its account and ARR, producing scored, revenue-ranked opportunities. A read-first MCP server brings all of this into AI agents (Claude Code, Cursor, VS Code, Windsurf, Slack).

For an expanded, single-file copy of everything below — full feature list, every FAQ answer, the complete MCP tool reference with an example call/response, and the glossary — see https://feedbackgraph.com/llms-full.txt

## Product

- Two-click capture widget (under 40KB gzipped) that attaches console logs, network requests, JS errors, screenshots, and screen recordings to every report.
- AI enrichment that auto-writes a title, summary, type, and severity, shown beside the raw evidence it was derived from — so engineers can trust the verdict instead of taking it on faith.
- Semantic dedupe (pgvector) that groups near-identical reports into one issue with an occurrence count.
- Revenue-backed Evidence Graph: every signal rolls up to its account and ARR, producing scored, revenue-ranked opportunities and opportunity dossiers.
- Multi-tenant security: role-based access control (6 roles) enforced server-side, Postgres row-level security, AES-256-GCM encrypted credentials, PII redaction, and GDPR export/delete.
- Links: Features https://feedbackgraph.com/features · Use cases https://feedbackgraph.com/use-cases · Compare https://feedbackgraph.com/compare

## How it works

1. Add one JavaScript snippet to your site (or share a hosted feedback URL — no code).
2. A user reports a bug; FeedbackGraph redacts private data, then AI writes the title, severity, and summary with the evidence attached.
3. The issue routes to your tracker, the right people are notified, and the reporter hears the moment it ships.
- Quickstart: https://feedbackgraph.com/docs
- Interactive demo (no login): a 7-step, click-through walkthrough of the whole loop — one-click capture, the AI verdict shown beside its evidence, semantic dedupe, one-click routing with two-way sync, the closed loop, and revenue-ranked opportunities — on a mock "Acme Inc" tenant: https://feedbackgraph.com/demo-tour

## MCP server

FeedbackGraph ships a tenant-scoped Model Context Protocol (MCP) server so an AI agent (Claude Code, Cursor, VS Code, Windsurf, or a Slack agent) can pull a report with full debugging evidence, search reports and signals, and read the revenue-ranked opportunities — without leaving the agent. It is read-first with one flag-gated write action, reuses the same authorized read paths as the dashboard, and is tenant-isolated by Postgres row-level security.

- Overview (for AI agents): https://feedbackgraph.com/mcp-server
- Developer + AI-agent docs: https://feedbackgraph.com/docs/mcp
- Machine-readable manifest: https://feedbackgraph.com/.well-known/mcp.json
- Endpoint (streamable HTTP): https://feedbackgraph.com/api/mcp
- Auth: Authorization: Bearer <sk_live_ tenant API token> (create one in Settings -> API keys). Not OAuth.
- Transport: MCP streamable HTTP (POST JSON-RPC; Accept: application/json, text/event-stream). Local editors connect over stdio via the mcp-remote bridge.
- Tools: fg_list_reports, fg_get_report, fg_search (scope reports:read); fg_list_opportunities, fg_get_opportunity (scope intel:read); fg_route_report (scope reports:write, flag-gated).
- Resources: report://{id}, opportunity://{id}.
- Security: one tenant per token, scoped + revocable keys, PII-redacted evidence, per-token rate limits, audit-logged, fail-closed on bad scope or cross-tenant id.

## Agent discovery & machine access

Every discovery document below is live; start at the API catalog:

- API catalog (RFC 9727, linkset+json): https://feedbackgraph.com/.well-known/api-catalog
- OpenAPI 3.1 for the public API (includes MPP x-payment-info payment discovery): https://feedbackgraph.com/openapi.json
- Markdown for agents: request the homepage or any marketing/docs page with `Accept: text/markdown` to receive a text/markdown rendition (HTML stays the default; the response carries x-markdown-tokens).
- Agent registration + auth recipe: https://feedbackgraph.com/auth.md
- OAuth discovery (RFC 8414 + RFC 9728; client_credentials exchanges an sk_live_ API key): https://feedbackgraph.com/.well-known/oauth-authorization-server and https://feedbackgraph.com/.well-known/oauth-protected-resource
- Agent Skills index (sha256-pinned SKILL.md files): https://feedbackgraph.com/.well-known/agent-skills/index.json
- WebMCP: marketing pages register in-browser tools (doc search, pricing, MCP connect info, navigation) via navigator.modelContext on page load.
- Commerce discovery: https://feedbackgraph.com/.well-known/ucp · https://feedbackgraph.com/.well-known/acp.json · https://feedbackgraph.com/.well-known/x402
- robots.txt declares Content Signals (search=yes, ai-input=yes, ai-train=no); homepage responses carry RFC 8288 Link headers pointing at the catalog, spec, and docs.

## Integrations

- One-click and rules-based routing with two-way status sync to Jira, Linear, GitHub, and Slack, plus a signed generic webhook.
- More providers in progress: ClickUp, Monday.com, Asana, Trello, TestRail, GitLab, Azure DevOps.
- Closed-loop reporter notifications fire when a routed issue's status changes upstream.
- Details: https://feedbackgraph.com/integrations

## Pricing

Pricing is regional and detected automatically by location. Prices display in USD, EUR, AED, SAR or INR; billing runs on Polar (Merchant of Record) — India is charged in INR (UPI and cards), everyone else in USD.

Plans (USD monthly · INR monthly):
- Free Audit — $0 / ₹0 for 30 days, no card (~500 signals).
- Starter — $4.99/mo · ₹472/mo.
- Growth — $14.99/mo · ₹1,419/mo (most popular; full AI + all integrations).
- Pro — $39.99/mo · ₹3,787/mo.
- Scale — $79.99/mo · ₹7,574/mo.
- Enterprise — custom volume, security review, and onboarding.

Pay yearly and get 2 months free (~17% off) — one upfront charge of 10× the monthly rate. Subscriptions auto-renew; cancel anytime from the billing page.
- Pricing page: https://feedbackgraph.com/pricing

## Docs

- Quickstart (install the widget in minutes): https://feedbackgraph.com/docs
- MCP server docs (tokens, stdio + HTTP setup, tool reference, scopes, security): https://feedbackgraph.com/docs/mcp
- Roadmap (shipped, in progress, next): https://feedbackgraph.com/roadmap
- Book a demo: https://feedbackgraph.com/demo
- Contact sales: https://feedbackgraph.com/contact

## FAQ

Full answers: https://feedbackgraph.com/faq

- Is the AI a black box? No. Every AI verdict is shown next to the console errors, network requests, JS exceptions, and on-screen text it was based on.
- Do I need a credit card to start? No. The Free Audit runs for 30 days with no card.
- Where does feedback route to? Jira, Linear, GitHub, and Slack with two-way status sync, plus more providers and a signed webhook.
- How is my data protected? PII is redacted before AI sees it; tenants are isolated by Postgres row-level security; credentials are encrypted; GDPR export/delete is self-serve.
- Does the MCP server cost extra? No. It reuses your existing plan's read paths with scoped tokens.
- How does it increase revenue? Every bug and feature request is ranked by the accounts and ARR behind it, so teams fix and build what protects renewals and drives expansion.
- Does it help QA and UAT? Yes — testers file reproducible two-click bug reports with full technical context, triage on a shared board, and sign off UAT faster, catching regressions before release.
- How do I catch bugs before customers? Capture during QA, UAT and dogfooding with console/network/error evidence, let AI flag severity, and route regressions to your tracker in one click for more stable releases.

## Glossary

Definitions: https://feedbackgraph.com/glossary

- Evidence Graph — the revenue-backed graph linking every signal to its account and ARR to rank opportunities.
- Revenue-backed opportunity — a scored theme ranked by the accounts and revenue behind it.
- Closed-loop — automatically telling the original reporter when their issue changes status or ships.
- MCP server — a Model Context Protocol endpoint that lets AI agents read FeedbackGraph reports, evidence, and opportunities.
- How we compare to Jam.dev, Canny, and Enterpret: https://feedbackgraph.com/compare