The MCP tool directory.

Get the accessibility tree for any URL without needing a browser session. Returns a structured snapshot of all interactive elements, headings, links, buttons, form fields, images with alt text, and ARIA roles. Great for quick UX audits.

[write] Write Facebook/Instagram ad copy for an ad image URL: headline, primary text, description, CTA, and hook angle. Returns up to 5 variations. Pass projectId so the copy uses the brand brief. Call this right after a generation completes.

Get a secure one-time link to register a new database connection with ThinAir Data (postgres, mysql, or mssql). This tool does NOT take a connection string as input — you'll open the returned link and paste the connection string into a secure web form; it is never sent through chat. The response includes `connection_string_format` and `auth_note` for the chosen dialect — surface both to the user verbatim. IMPORTANT for mssql: Azure SQL uses Microsoft Entra, so the connection string is HOST/DATABASE only (no credentials) and the tenant/client/secret go in the form's separate fields — never construct or suggest an `mssql://CLIENT_ID:CLIENT_SECRET@host` string (client secrets break URL parsing).

[write] Queue a FINISHED AD image render. Required: projectId. Call adResearch.listProjects first and pick the project the request is about, or ask the user when it is ambiguous; never guess an id. Also required: prompt, a creative brief rather than a finished image prompt. How the prompt is used: a creative-director model rewrites it into the real image prompt using the project's brand brief, and that rewrite always adds a specific headline, a subline, a CTA button, and the project logo when the project has one. Negative instructions in the prompt ('no text', 'no logo', 'nothing but the product') cannot suppress those; use adGeneration.generateImage when the caller wants exact or no text. Optional: adType, aspectRatio, quality (low/high/ultra), referenceImageUrls, inputImageId (from upload.uploadImageBase64), referenceAdId or referenceInspirationId to remix an existing ad. Costs credits. Returns a generation ID. Poll adGeneration.getById until status is COMPLETED.

[write] Queue an image-to-image edit and return immediately. The universal edit tool: change background, restyle, swap a subject, change aspect ratio, add brand assets, or remix a competitor ad. Source is either sourceAdId (one of the user's own generations) or sourceImageUrl (any direct image URL). Costs credits. Poll adGeneration.getById until COMPLETED.

[write] Queue an image render from a VERBATIM prompt. The prompt goes to the image model exactly as written: no creative-director rewrite, no brand brief, no added headline, subline or CTA, and no project logo. The image contains only what the prompt stipulates. Use this when the caller wants only specific stipulated text, forbids text/logos/CTAs, or wants raw scene, product, or photography imagery rather than a finished ad; use adGeneration.create for finished ads. Required: projectId (credit scoping only, nothing from it enters the prompt) and prompt. Optional: aspectRatio, quality (low/high/ultra), referenceImageUrls and inputImageId for image-to-image. Costs the same credits as adGeneration.create. Returns a generation ID. Poll adGeneration.getById until status is COMPLETED.

Get the status, prompt, and resulting image URLs of a single ad generation by ID. Use to poll until status is COMPLETED.

[write] Add competitors to a project by name (optionally with website URL) and kick off ad scraping for them.

Get one scraped competitor ad in full: creative URLs, ad text, headline, run dates, and platform placements.

Download a file from Agent Drive. Returns a signed download URL valid for 1 hour.

Download an email attachment. Returns a signed download URL (valid 1 hour), filename, MIME type, and size. Use email_read first to see available attachments and their indexes.

Fetch raw CSV datasets and investigation metadata. Returns dataset URLs, lens structures, key correlations, open questions, and outliers for independent analysis.

Upload an attachment by providing either a file or a valid URL. Maximum request body size: 100MB.

Upload an attachment by providing either a file or a valid URL. Maximum request body size: 100MB.

Upload an attachment by providing either a file or a valid URL. Maximum request body size: 100MB.

Upload an attachment by providing either a file or a valid URL. Maximum request body size: 100MB.

Get an Audit Log Export. The URL will expire after 10 minutes. If the export is needed again at a later time, refetching the export will regenerate the URL.

One-time setup: Connect the user's Gmail account via OAuth so the AI can read verification emails automatically. Returns an authorization URL the user must visit. After authorizing, the AI can use read_verification_email to fetch OTP codes.

Performs a batch analysis of multiple URLs, domains, or subdomains to retrieve selected SEO, backlink, organic, and paid traffic metrics. When analyzing a domain name, you must use mode=subdomains. Using mode=domain can exclude www and other subdomains. Use doc tool first to get the real input schema.

Provides the number of impressions for your and competitors's brands in an LLM you specify, with filters for locations, query text, URL, and more. Use doc tool first to get the real input schema.

Provides the number of mentions for your and competitors's brands in an LLM you specify, with filters for locations, query text, URL, and more. Use doc tool first to get the real input schema.

Provides the share of voice for your and competitors's brands in an LLM you specify, with filters for locations, query text, URL, and more. Use doc tool first to get the real input schema.

Retrieves pre-extracted, relevance-ranked web content using Brave's LLM Context API, optimized for AI agents, LLM grounding, and RAG pipelines. Unlike a traditional web search that returns links and short descriptions, this tool returns the actual substance of matching pages — text chunks, tables, code blocks, and structured data — so the model can reason over it directly. When relaying results in markdown-supporting environments, cite source URLs from the `sources` map.

Close the browser session and free all resources. Always call this when the browser workflow is complete. If the session was started with record_video: true, the video recording URL is returned.

Open a browser and navigate to a URL. Returns a screenshot of the loaded page. Use this to start a browser session — the returned sessionId must be passed to all subsequent browser_ tools. Pass width/height to start with a custom viewport (e.g. 393×852 for mobile). Set record_video to true to record the entire session as a video — the recording URL is returned when browser_close is called. When workflow metadata is provided, the resulting run can surface structured verdicts, summaries, and next actions in the dashboard.

Get the full network request waterfall with timing data. Shows every request made by the page — URLs, methods, status codes, resource types, durations, and sizes. Use for performance analysis and debugging.

Extract SEO metadata from the current page: title, meta description, Open Graph tags, Twitter cards, canonical URL, heading hierarchy, structured data (JSON-LD), robots directives, and image alt text coverage.

Build an itemized CAD quote (motor + installation + propeller + trade-in + HST + financing tier) and return a deep-link URL the customer can open. Optional contact captures a lead in the dealership CRM.

Search the Ceramic AI knowledge base for relevant information. Returns ranked results with titles, URLs, and descriptions.

Upload an audio or speech file to detect whether it is human-recorded or AI-synthesized. Provide the audio content as a base64-encoded string. Returns a classification identifier for async result retrieval. WARNING: base64 encoding adds ~33% overhead to the original file size. For audio files larger than 10 MB, use classify_audio_url instead and provide a publicly accessible URL to avoid payload size issues. Authentication: provide your Identifai API key via the apiKey parameter or configure the X-Api-Key HTTP header in your MCP client (recommended).