Kind: mcp-server

Use when an agent needs ALMA scorecard data - regions, scorecards, indicators, period-to-date maps, data-point time series and region geo for a given ALMA tenant. Wraps ALMA's embedded, read-only hosted MCP endpoint (per-tenant `/mcp` on the tenant API host), so every tool runs as the connected Personal Access Token's role + region scope. Reach for this over raw REST/OpenAPI when the task spans several ALMA reads or you want typed, permission-scoped tools (e.g. joining indicator time series against weather).

A stateful, formula-aware Excel engine an agent can actually drive - not a file it has to parse by hand. Opens existing .xlsx/.xlsm/.xls workbooks (or creates new ones) entirely in memory, then queries and richly edits them across a whole session without reloading: cell values, typed formulas, styling, conditional formatting, data validation, charts, merged regions, cell notes and threaded comments, sheet structure and outline grouping, rows and columns, and named ranges (tables are read-only) - flushed to disk with an atomic, corruption-safe save. Its standout capability is headless recalculation: Apache POI evaluates ~280 Excel functions in place, so the agent works with real computed results instead of the stale cached zeros that code-based approaches (openpyxl/pandas, or Claude's built-in spreadsheet handling) leave behind - and every read distinguishes a genuine value from an as-yet-uncomputed formula. It can also review a model, not just write one: workbook.audit scans for hardcoded constants and hidden rows, and cell.trace walks precedents and dependents. Because all edits flow through one POI writer, it avoids the 'Excel repaired records' corruption a second serializer over the same file causes. A workbook too large to load can still be read: workbook.open's read-only stream mode pages through it in flat memory. 55 tools across workbook lifecycle (incl. audit and stream mode), agent context, range I/O (incl. cell styling, conditional formats written and read back, data validation, formula tracing, formula-pattern search, and cell comments), sheet management (incl. merge/unmerge, sheet reordering and row paging), presentation written and read back, print setup and outlining, line and bar charts, tables, named ranges, read-only VBA extraction, and multi-call batching.

Use when an agent needs to act on GitHub - read or open issues and pull requests, inspect or edit repository contents, trigger or read GitHub Actions, or run code/issue/PR search across repos the connected identity can reach. Wraps GitHub's official hosted remote MCP server (api.githubcopilot.com/mcp), so tools are scoped to the user's GitHub permissions. Reach for this over raw REST/GraphQL when the task spans several GitHub operations or you want typed, permission-scoped tools rather than hand-rolled API calls.

Use when an agent needs to author, modify, review, render, or export PowerPoint (.pptx / .pptm) decks — create/open/close documents, add/duplicate/delete/reorder slides, edit text (with run/paragraph/shape scope + styling), insert images with picture effects, add/edit tables (incl. merge + per-cell borders), draw shapes (solid/gradient/pattern fills), update chart data in existing charts, render slides to PNG/JPG/SVG at low or high fidelity, and export to PPTX/PDF/HTML/PNG-batch/JPG-batch/SVG-batch/outline-text. Stateful Apache POI 5.5.x + LibreOffice engine over stdio, 52 tools.

A dialect-agnostic SQL toolset giving an agent permission-scoped access to PostgreSQL, MySQL, SQLite, and MSSQL through one unified set of tools. Introspect schemas/tables/columns, run parameterized read-only SELECTs with stateless keyset pagination, and (in dml/full scopes) execute parameterized writes and atomic batches — all behind a statement classifier that blocks statement stacking, comment evasion, and data-modifying CTEs, plus DB-level read-only enforcement. One database (DSN) per server instance; the dialect is selected by config.

Use when an agent needs the connected athlete's own Strava data - list recent or historical activities, inspect a single activity's details or time-series streams, read athlete profile and season stats, explore segments and segment efforts, or export a saved route as GPX/TCX. Runs as a local stdio server against Strava's REST API (api/v3) using credentials the user supplies, so every call is scoped to that one athlete's account. Reach for this over raw HTTP when the task spans several Strava reads or needs typed tools rather than hand-rolled API calls. Personal-scope only: Strava's API terms allow an athlete's data to be shown to that athlete alone, so this asset must never be assigned at project or organization scope.

A stateful Word (.docx) engine for source-free document creation, Template composition, and editing a real document in place instead of regenerating it. Its 38 tools expose a neutral blank-package baseline; identity-bound Blocks across body, headers, footers and text boxes; Named styles and resolved style audits; scoped text, paragraph and table composition; live PDF previews; core properties and Word fields; PNG/JPEG replacement and insertion; and complete Revision/Comment collaboration. document.save backs up, verifies the OOXML Part inventory and content, and atomically replaces the target. Use it when creating a new document or preserving a Template, numbering, review markup, media and package fidelity.