Complete coverage of the three Google Docs API v1 operations through six direct, bounded methods under per-file drive.file access. The model chooses the exact read or action. Reads are response-capped and tab-aware; writes use closed schemas and server-computed field masks. Raw batchUpdate is not exposed as an unchecked request escape hatch.
6 methodsFreshness not declaredUnified OneDrive + SharePoint document ingestion and structured extraction. CRITICAL USAGE RULES: 1. Use ingest_onedrive_folder() to list/search OneDrive files, download, parse, and normalize. 2. Use ingest_sharepoint_library() to resolve site/drive, ingest library files, and normalize. 3. Parsers support: PDF, CSV, XLSX/XLS, DOCX/DOC, TXT (max file size configurable). 4. Incremental mode supports existing itemId + etag/lastModified to skip unchanged files. 5. Use extract_structured_data() for strict typed extraction (invoice, lease, report). 6. Use upsert_document_record() to keep raw text and extracted JSON separate in a deterministic record shape.
5 methodsFreshness not declaredExact Google Drive API v3 file and permission operations. The model chooses one named operation and supplies its exact arguments; the connector does not semantically route requests or create hidden intermediate folders. The OAuth drive.file scope limits discovery to files PatchOps created or the user explicitly opened with PatchOps. Downloads return an artifact marker instead of placing file bytes in model context. Workspace files require an explicit exportMimeType. deleteFile permanently deletes; it does not move files to Trash.
11 methodsFreshness not declaredMicrosoft Excel connector for workbook automation via Microsoft Graph. CRITICAL USAGE RULES: 1. Provide itemId (drive item ID for the workbook) for all calls. 2. Use siteId when working with SharePoint-hosted workbooks. 3. For long-running workflows, create a session and pass workbook-session-id on subsequent calls. 4. Range addresses should use Excel A1 notation (e.g., Sheet1!A1:B10).
82 methodsFreshness not declaredGitHub connector for managing repositories, issues, pull requests, commits, and code. CRITICAL USAGE RULES: 1. Use listRepositories() to discover available repos before other operations. 2. Repository names use owner/repo format (e.g., "anthropics/claude-code"). 3. Use searchCode() for finding code patterns across repositories. 4. Issue and PR numbers are integers, not IDs. 5. File contents are returned as base64 - decode if needed. AUTHENTICATION: - Uses OAuth token from user's connected GitHub account - Scopes: repo, read:user, user:email, read:org
18 methodsFreshness not declaredGitHub Projects V2 connector for managing project boards, items, fields, and views. CRITICAL USAGE RULES: 1. Use listProjects() to discover project IDs before other operations. With no args, it lists the authenticated user's projects. 2. Projects use numeric IDs (project_number) visible in the URL. 3. Items can be issues, PRs, or draft issues. 4. Use GraphQL field IDs from listProjectFields() for field updates. 5. View filters and sorts use GraphQL syntax. AUTHENTICATION: - Uses OAuth token from user's connected GitHub account - Scopes: repo, read:user, user:email, read:org, project
15 methodsFreshness not declaredExact Gmail API v1 operations authorized by the connected gmail.modify grant. The model chooses one direct method. Reads are bounded and raw RFC 2822 base64 is not exposed. Attachments return artifact markers. Sends and drafts use CR/LF-safe MIME construction. Trash is reversible and untrash is exposed. Permanent delete is intentionally absent because Google's messages.delete requires the broader mail.google.com scope that PatchOps does not request.
50 methodsFreshness not declaredTwenty-six exact Google Calendar API v3 methods over one serializer. The model selects the operation; PatchOps validates and serializes it without semantic routing. Coverage is deliberately reductive: webhook lifecycle calls, provider-side natural-language quick-add, catastrophic calendar clearing, admin ownership transfer, and unsafe full-replacement duplicates are explicitly excluded. Reads and writes are bounded, mutations use closed schemas, and guest-affecting event changes require an explicit sendUpdates policy.
26 methodsFreshness not declaredComplete current HubSpot public API catalog: 1,246 callable operations across 117 pinned official specs. Methods are lazily discovered with exact closed inputs; write operations require action approval. OAuth token issuance, provider test/debug routes, and callback-completion internals are deliberately excluded.
1246 methodsFreshness not declaredMicrosoft OneDrive connector for listing, searching, downloading, and uploading files. CRITICAL USAGE RULES: 1. Use listDriveItems() first to discover folder and file IDs. 2. downloadDriveItem() stores bounded bytes behind a private short-lived download link; do not inline file bytes. 3. searchDriveItems() searches from the drive root only. 4. uploadDriveItem() requires filename + exactly one of contentBase64, contentText, or sourceUrl. Prefer sourceUrl (server fetches the bytes) over contentBase64 for large files. 5. listDriveItems()/searchDriveItems() support select/top and fetchAll for pagination. 6. getItemByPath(), createFolder(), deleteItem(), moveItem(), copyItem(), and createShareLink() are explicit path and folder operations.
10 methodsFreshness not declaredOutlook mail: search and read messages, parse attachments, send email with files. RULES: 1. outlookMail_get_and_parse_emails retrieves and parses in one step; or get_emails_with_attachments, then parse_email_attachments. 2. Attachments up to 30 MB per file and message; files over 3 MB go through a draft + upload session (saves to Sent Items). 3. outlookMail_send_email_with_files sends generated files (contentText, contentBase64, or path). 4. Raw downloads default to metadata; mode:"inline" only for small files. 5. Multi-criteria search: outlookMail_advanced_search. 6. Archive = folderId:"archive"; Online Archive is unsupported. 7. Write plainly: short sentences, active voice, everyday words. Lead with the ask; keep the user's wording. Plain text keeps line breaks and "- " lists.
44 methodsFreshness not declaredMicrosoft Outlook calendar connector for calendars and events. CRITICAL USAGE RULES: 1. Use outlookCalendar_list_calendars() to discover calendar IDs before event operations. 2. outlookCalendar_list_events() supports documented OData selection, filtering, ordering, and bounded pagination. 3. Omit calendarId for the default calendar; provide an exact calendarId for another available calendar. 4. Event IDs and calendar IDs are opaque values discovered through read methods.
7 methodsFreshness not declaredRetrieve Outlook email and parse attachments. Use getEmailsWithAttachments for metadata, parseEmailAttachments for one message, or getAndParseEmails for both. DOCX, XLSX, XLS, CSV, and TXT become Markdown; PDFs and images return in __PATCHOPS_FILES__ as native model content. Legacy .doc is unsupported. fileTypes filters extensions. Default max size is 10 MB; adjust maxFileSizeMB. DOCX tables render as HTML.
4 methodsFreshness not declaredMicrosoft Planner connector for managing plans, buckets, tasks, and task metadata. CRITICAL USAGE RULES: 1. Use planner_list_plans() to discover plan IDs before task or bucket operations. 2. Updating tasks requires the latest ETag; the planner_update_* helpers auto-fetch when omitted. 3. Attach only an existing HTTPS reference; file upload belongs to a file-storage connector.
14 methodsFreshness not declaredMicrosoft Power BI REST API. The model selects one exact Microsoft operation; PatchOps validates and serializes it without intent routing. READS include workspaces, dashboards, reports, datasets, DAX executeQueries, refresh history, gateways, imports, capacities, and admin inventory. ACTIONS include dataset/report/dashboard/workspace management, refreshes, pipelines, imports, embed tokens, gateways, scorecards, and admin operations. Binary report exports return download handles rather than inline bytes.
287 methodsFreshness not declaredExact Microsoft Graph file operations for .ppt and .pptx presentations stored in OneDrive or SharePoint. CRITICAL USAGE RULES: 1. Microsoft Graph does not expose slide editing through these methods; this connector operates on presentation files. 2. Use list or search to discover opaque item IDs. The model chooses the exact method; there is no semantic router. 3. Download and PDF export return artifact markers so file bytes do not enter model context. 4. Upload accepts one exact target and one source. Prefer sourceUrl over inline base64. 5. Preview returns only Graph's documented getUrl, postUrl, and postParameters fields.
7 methodsFreshness not declaredMicrosoft SharePoint connector for sites, lists, drives, and permissions. CRITICAL USAGE RULES: 1. Use sharepoint_search_sites() or sharepoint_get_site_by_path() to resolve site IDs. 2. Use sharepoint_list_lists() to discover list IDs before item/column operations. 3. Use sharepoint_list_drives() to locate drive IDs for document libraries. 4. upload/download file content use base64 payloads for binary data. 5. For bulk operations, use sharepoint_bulk_upload() and sharepoint_bulk_download(); they use bounded direct file calls. 6. For path-based operations, use sharepoint_upload_by_path() and sharepoint_download_by_path().
60 methodsFreshness not declaredEight direct, bounded Google Sheets API v4 primitives authorized by drive.file for spreadsheets PatchOps created or the user explicitly opened. The model selects the method and supplies exact spreadsheet IDs and A1 ranges. Reads are response-capped; writes use closed schemas, bounded cell counts, and bounded JSON bodies. Generic batchUpdate is not exposed as an unchecked escape hatch.
18 methodsFreshness not declaredComplete coverage of the five Google Slides API v1 operations through eight direct, bounded methods under per-file drive.file access. The model chooses the exact read or action. Reads are response-capped and include thumbnail discovery; writes use closed schemas and explicit operations. Raw batchUpdate is not exposed as an unchecked request escape hatch.
8 methodsFreshness not declaredMicrosoft Teams connector for teams, channels, members, chats, and messages. RULES: 1. Use list_teams() or get_team_by_name() to resolve team IDs before channel/member actions. 2. Provide teamId whenever possible to avoid name collisions. 3. Channel operations need channelId or channelName plus teamId. 4. send_chat_message() takes chatId, email, or userId; email/userId resolves or creates the chat and sends in one action. 5. get_chat_id_with() returns a chatId string and creates the chat if needed. 6. Write messages plainly: short sentences, active voice, everyday words, no filler. Lead with the ask. Keep the user's wording. PARAMETER ALIASES: - search_users(), search_messages(): "query" (also searchTerm/search/q) - get_chat_id_with() accepts both "email" and "userId"
26 methodsFreshness not declared
