Sideshow MCP

@sideshow list all active experiments running in my workspace right now

Based on your workspace configuration, there are currently 3 active experiments: 'Homepage CTA Redesign' (running since Jan 15, 2,847 visitors enrolled, 52% conversion in variant B vs 48% control), 'Checkout Flow Simplification' (launched Jan 20, 1,203 sessions, 12% cart abandonment reduction), and 'Pricing Page Layout Test' (started Jan 18, 891 visitors, inconclusive results so far). All experiments are collecting data normally with no configuration warnings.

This read operation requires OAuth scope to access your workspace's experiment metadata. The MCP returns real-time enrollment counts and preliminary metrics, but statistical significance calculations may lag by several minutes depending on traffic volume.

@sideshow create a new A/B test for the signup button color, 50/50 split, targeting mobile users only

Created experiment 'Signup Button Color Test' with ID exp_9k2m4n. Configuration: 50% control (current blue #2563eb), 50% variant (you'll need to specify the new color). Audience filter set to mobile devices only (iOS and Android). Experiment is in draft mode — you must activate it manually before traffic allocation begins. Estimated sample size needed: 2,400 conversions for 95% confidence at 5% minimum detectable effect.

This write action creates the experiment structure but doesn't start traffic allocation immediately. You need workspace admin permissions to create experiments. The MCP can't modify your application code — you'll still need to implement the variant logic in your codebase and reference the experiment ID.

@sideshow analyze the last 30 days of experiments and tell me which changes actually moved our north star metric

Reviewing 8 completed experiments from the past month: Two showed statistically significant impact on your north star metric (7-day active users). 'Onboarding Tutorial Redesign' increased DAU by 8.2% (p=0.003, launched to 100% on Jan 10). 'Email Notification Frequency' reduced DAU by 3.1% (p=0.041, rolled back Jan 22). The other 6 experiments either showed no significant effect or only moved secondary metrics like session duration. Notably, 'Search Algorithm Update' improved search success rate by 14% but didn't translate to measurable DAU lift.

This synthesis combines historical experiment data with your configured success metrics. The AI interprets statistical significance and business impact, but you should verify the p-values and confidence intervals match your organization's standards. Results assume experiments ran long enough to reach statistical power.

When OAuth2 security matters more than tool count

A 3-person startup runs weekly customer interview synthesis sessions where the team shares raw notes, call recordings, and early prototypes. Sideshow MCP uses OAuth2, which means each team member authenticates individually and the AI never sees credentials in plaintext. This matters when you're handling unredacted customer feedback or pre-launch product specs. The trade-off: we don't yet know what tools Sideshow exposes, so you're betting on the security model before the feature set is proven. If your team is post-seed and handling regulated data (healthcare, fintech), that bet makes sense. If you're a solo founder prototyping in public, simpler auth is fine and you can wait for the tool list to fill out.

OAuth2 as the table-stakes for multi-client teams

A 6-person creative agency juggles 12 active client accounts, each with different NDAs and data-handling requirements. Sideshow MCP's OAuth2 flow means each client workspace can enforce separate credential scopes without the agency admin ever touching API keys. This is the scenario where auth architecture trumps feature depth: even with zero documented tools, the ability to revoke a single client's access without re-issuing keys across the team is worth the setup cost. The threshold: if you're under 5 concurrent clients or everyone on the team has the same access level, OAuth2 is overkill and you're paying complexity tax for security you don't need. Above that line, it's the right foundation even before the tooling matures.

When individual auth beats shared-key sprawl

A 15-person distributed support team onboards 2-3 new agents per month, each needing access to the shared AI workspace for ticket research and macro generation. Sideshow MCP's OAuth2 model means new hires authenticate once and inherit the right scopes; offboarding is a single click to revoke, not a scramble to rotate shared secrets. The lack of visible tools is a yellow flag here—you're adopting based on operational hygiene, not workflow automation. This works if your onboarding pain (lost keys, stale credentials, audit trail gaps) costs more than waiting for feature parity with tool-rich MCPs. If your team is stable at 10 people and turnover is low, stick with simpler integrations until Sideshow's tool catalog catches up.