FSFeature State
Registry 001V0.1

Independent capability registry · v0.1

Last research pass

Know what AI
actually supports.

Product. Plan. Platform. Evidence.Feature State tracks the conditions that turn a tempting “yes” into a useful answer.

Claim“ChatGPT supports MCP”
StateSelected plans
ScopeBusiness · Enterprise · Edu
Evidence1 primary source
Last verified

Search the registry

Verified July 27, 2026

010products
015capabilities
150states recorded
114source-backed

01 / Live matrix

Six products.
Four decisive states.

A dot never means “yes everywhere.” Open any state to see the plan, platform, conditions, and evidence.

Explore all 150 recorded states 114 currently have primary evidence. Unknown records are research gaps, not negative claims.
ProductMCP clientBrowser controlTerminal accessBackground agents
01ChatGPTPlansPlansUnknownPlans
02ClaudeYesPlansUnknownPlans
03PerplexityPlansPlansUnknownPlans
04CodexYesBetaYesYes
05CursorYesYesYesYes
06GitHub CopilotYesConditionalYesPlans
YesPlansConditionalBetaUnknown

03 / Change record

Capability states
have a history.

We keep the previous state, new state, date, and source instead of quietly rewriting a table.

  • CodexPersistent memoryUnknownConditionalAll plans3 primary
  • Replit AgentImage generationUnknownConditionalProduct-wide2 primary
  • GitHub CopilotDesktop applicationUnknownYesmacOS, Windows, Linux2 primary
  • GitHub CopilotPersistent memoryUnknownPlansPro, Pro Plus2 primary
  • Replit AgentPDF analysisUnknownYesAll plans2 primary
Read the public change log

04 / Research standard

“Unknown” is more honest than a guessed no.

Feature State separates missing evidence from unsupported behavior. We capture product surface, plan, platform, source, confidence, and access date behind every classified state.

  1. 01
    Find primary evidence

    Official docs, support pages, release notes, and repositories come first.

  2. 02
    Scope the exact claim

    Product, plan, platform, region, and rollout are kept separate.

  3. 03
    Classify confidence

    Evidence quality and ambiguity determine high, medium, or low confidence.

  4. 04
    Preserve the change

    Updates append history instead of erasing the prior state.

  5. 05
    Reverify on schedule

    Material vendor changes trigger a new source review and dated record.

Read the full methodology

05 / Start with a comparison

Same question.
Two products.

No universal
winner
01

ChatGPT vs Claude

MCP clientPlansvsYes
Browser controlPlansvsPlans
View comparison
02

Codex vs Claude Code

Browser controlBetavsPlans
Terminal accessYesvsYes
View comparison
03

Cursor vs GitHub Copilot

Browser controlYesvsConditional
Background agentsYesvsPlans
View comparison

Researched and reviewed by . Primary evidence preferred; corrections remain public.