One profile, many platform records
Each tool (for example, a social media scheduler) has a single canonical profile. Its support for each AI assistant — ChatGPT, Claude, Gemini, Copilot and others — is stored as a separate platform record with its own status, method, capabilities, sources and verification date.
Listing status
- Official directory Appears in the AI platform's own directory, confirmed by a link to that listing.
- Vendor supported Documented and supported by the tool vendor, but not (or not yet) in the platform directory.
- Community built Built and maintained by third parties, not the vendor.
- Unverified Reported or suspected, without sufficient evidence.
Connection method is separate
How a connector works — Native Remote MCP API — is tracked independently of listing status. An official directory listing can use MCP. Conversely, the existence of an MCP server never implies official status on any platform.
Sources & verification dates
A record counts as verified only when it has at least one source URL (platform directory, vendor documentation, or a reputable community source) and a date on which a reviewer last checked it. Records without both are shown as not verified.
Imports & review
Directory imports land in a staging area first. Each candidate is de-duplicated against existing tools and approved or rejected by a human reviewer. Every change to tools, platform records and sources is written to a change log.
Demo data
While the directory is being built, some records are illustrative examples marked Demo · Unverified. They are not compatibility claims and are excluded from search indexing.
Affiliate disclosure
Some outbound links may in future be affiliate links, meaning ConnectorFire could earn a commission if you sign up. Affiliate relationships never influence listing status, verification or ordering. Sponsored placements, if any, will be clearly labelled.
AI visibility measurement
Catalog presence is not an AI visibility measurement. All tools start as Not tested. No scores, rankings or percentages are created without imported test evidence. Mention means the tool was named; recommendation means it was suggested. Surfacing means an actual in-product connector was displayed; activation means an observable activation event, not a recommendation or an assumed install.
Every eligible vendor or integration must have a row for every trial, including explicit false results when it was not mentioned or recommended. Several vendors may appear in the same trial. Blank metrics mean not measured, never zero. Eligible completed trials form the sample count; each rate uses only eligible completed trials where that specific metric was measured. Failed and unmeasured trials are excluded, not silently counted as successes or negatives. Evidence references are required for completed trials.
API simulations, first-party UI observations and other observed tests are separate methodologies. API output cannot establish in-product surfacing or activation. Exact assistant, consumer or enterprise edition, model/version, product plan, question/version, sampling configuration and dates remain attached to results. Gemini consumer and enterprise are separate; Microsoft 365 Copilot and Copilot Studio / Power Platform are separate. No compatibility is inferred from visibility tests.
Category comparisons order only tested recommendation samples from the same run, question version and exact set of measured trial IDs, separately for vendor-level and integration-level targets. Untested tools are not ranked. Counts include zero recommendations. Displayed 95% Wilson binomial confidence intervals describe sampling uncertainty, not causal superiority; overlapping intervals do not establish a meaningful ranking difference.
Question selection, repeat prompts, geography, language, personalization, account history, enabled connectors, product plan, availability, model changes and time can bias results. Repeated or correlated trials violate independent-binomial assumptions and can make intervals too narrow. Samples may not generalize to other users or plans. Run-specific sampling and limitations are shown with results. Do not compare across editions or methodologies or interpret unmeasured activation as failure.
Tests are manually supplied and evidence-reviewed. ConnectorFire does not autonomously scrape product interfaces, run unauthorized tests or bypass platform terms. Raw observations are administrator-only; aggregate evidence references must be cleared for publication. Imports and repeatable aggregation never change catalog compatibility or official verification labels.