Tools

Official Spark mail skills: agents read mail, calendar, contacts and teams, with flows for morning briefs, inbox zero, meeting prep, follow-ups and weekly reports. Needs Spark desktop on macOS or Windows with the CLI enabled; set read-only, organize or send access per account.

Searches ZIM-format knowledge bases such as Wikipedia, Wiktionary and Stack Exchange offline: full-text search, read articles by section and follow links, even with no internet. Download .zim files from Kiwix first (hundreds of MB to tens of GB) into ~/zim-files, and install uv.

Octagon AI's official research MCP: ask in plain language about SEC filings, earnings calls, financial metrics, stock and crypto prices and news; also prediction-market reports and a deep research agent. Needs a free API key and Node.js. Not investment advice; verify findings.

Official VideoDB skills: agents ingest video from local files, URLs, YouTube or live streams, then transcribe, index, search clips by content, generate subtitles, edit and dub, returning playable links. Needs Python 3.9+ and a VideoDB API key (free credits on sign-up).

Official BibiGPT skill: hand agents YouTube, Bilibili, podcast or local audio and video to get summaries, timestamped transcripts, chapter notes and mind maps. Needs a BibiGPT account (free tier has daily limits) plus its desktop app (with the bibi CLI) or an API key.

Connects any FHIR-compliant EHR server (e.g. the Epic sandbox or HAPI) to agents to search, read, create and update patient clinical data in plain language, with SMART on FHIR auth. By WSO2. Needs a FHIR server URL, access token and uv; clinicians must review results.

Salesforce's official hosted MCP servers: after an admin enables them in Setup with an OAuth external client app, agents can query and update CRM records and run Flows under the user's permissions. Needs Enterprise Edition or higher, or a free Developer Edition org. Repo holds docs and Q&A.

Connects to the St. Louis Fed's FRED database: search and pull 800,000+ macroeconomic time series (GDP, unemployment, inflation, interest rates) with built-in year-over-year and period-over-period changes. For macro research, investing and teaching. Needs a free FRED API key and Node.js.

Lets agents view, create and update Zendesk tickets and comments, read Help Center articles as a knowledge base, and use built-in prompts for ticket analysis and reply drafting. Each support rep signs in with OAuth after an admin registers an OAuth client. Install from source.

Runs locally with no API key: looks up drug labels and adverse events from FDA, EMA and DailyMed, plus WHO health statistics, PubMed papers, clinical trials and clinical guidelines, with auto-generated evidence-level tags on papers. Needs Node.js. Have a doctor or pharmacist verify results.

Diffs a live page's accessibility violations against a baseline, stashing uncommitted changes by default or comparing with a branch, and reports new, fixed and pre-existing WCAG issues with selectors and fixes. Locates only, no edits. Runs AccessLint's Chrome and CLI tools via npx.

Runs the AccessLint rule engine against a live page and returns a worklist of WCAG 2.2 violations, each with its DOM selector, source file and line when mapped, evidence and a suggested fix. Locates without editing; sets up a debuggable Chrome and runs the CLI via npx.

Jenkins plugin that turns Jenkins itself into an MCP server: agents can list jobs, trigger and rerun builds, page through and search build logs, and view test results and SCM info. Hosted in the official jenkinsci organization; authenticates with a Jenkins API token over Basic auth.

Sonar's official plugin bringing SonarQube checks into the agent coding loop: secrets-scanning hook, issue fixing, quality gates, coverage and dependency risk; with Vortex analysis on, every edit is checked. Needs SonarQube (free Community Build works), the sonar CLI and Docker.

From Twilio Labs: exposes Twilio's public APIs (SMS, voice, phone numbers, Serverless and more) as MCP tools generated from its OpenAPI specs, with filtering by service or tag. Needs a Twilio API key; marked as a proof of concept.

Shapes the agent's own replies rather than your documents: separates what changed from what was verified, cuts invented confidence, leads with the action and keeps load-bearing caveats. Manually invoked and stays on until you say stop; a script can enable it in CLAUDE.md or AGENTS.md.

Neon's official plugin combining the Neon MCP and skills to manage serverless Postgres: projects and branches, object storage, Functions and the AI Gateway. Signs in to Neon via OAuth the first time you use the MCP.

Connects to PACS and other DICOM servers: query images by patient, study and series, read text from PDF reports embedded in DICOM, and forward series to AI nodes such as segmentation. Needs uv and a YAML config. Not for clinical use or real patient data (per the author).

Analyzes your codebase and generates a render.yaml Blueprint, or creates services directly through the Render MCP, to deploy apps to Render. Maintained by Render and included in OpenAI's curated skills. Needs a Render account and code pushed to GitHub, GitLab or Bitbucket.

Wolfram's official agent tools: use the Wolfram Language and Wolfram|Alpha for symbolic and numeric math and scientific and statistical data, for teaching and research. The command runs the official Docker image with a free developer license; Wolfram 15+ users can install it in the app.

Searches 600,000+ studies registered on ClinicalTrials.gov, shows protocols and posted results (efficacy, adverse events), and matches patients to recruiting trials by age, sex, condition and location. Public data, no API key; needs Node.js 24+. Unofficial; results must be reviewed by a doctor.

PostHog's official plugin with a built-in MCP and 30+ skills: query product analytics, feature flags, A/B experiments, dashboards and error tracking from chat, and optionally send Claude Code sessions to LLM Analytics. Needs OAuth sign-in with a PostHog account.

MLflow's official skills plugin: lets coding agents add MLflow Tracing to LLM apps, analyze failed traces, build evaluation datasets and scorers, and query token and latency metrics, closing the instrument, evaluate, iterate loop. Needs MLflow 3.8+.

Applies tiered prefers-reduced-motion handling instead of disabling everything: removes vestibular triggers like parallax and scroll hijacking, softens large movement into short fades, and keeps essential motion. Covers CSS and JS gating for GSAP, Framer Motion and Lenis, per WCAG 2.3.3.