From tools to useful work
Give your agent
a job worth doing.
Review a US forecast. Research a topic. Find a dataset. Compare a package. Check a launch. Start with the outcome, then connect the tools that can help.
Example workflows and prompts · review setup and permissions before connecting
Step-by-step MCP workflows
Research PubChem chemical identifiers and properties with AI and MCP
Resolve chemical names to compound candidates, then inspect exact CIDs, source identity, formula and calculated molecular weight.
Research UniProt protein records by gene and species with AI and MCP
Match a literal gene and exact organism, then inspect canonical protein accessions, source evidence, length metadata and versions.
Research U.S. Treasury debt records with AI and MCP
Inspect reported debt history and compare two exact publication dates with original decimal dollar amounts and explicit missing records.
Research NOAA coastal tide predictions with AI and MCP
Inspect coastal station metadata and bounded predicted high/low events with explicit GMT time, MLLW datum and height units.
Compare EU country statistics with AI and MCP
Research Eurostat unemployment, population and GDP with dated annual observations, explicit units, source flags and visible missing values.
Research ECB reference exchange rates with AI and MCP
Check published currency coverage, retrieve dated ECB reference rates and inspect labelled cross-rate and conversion calculations.
Research registered study records with AI and MCP
Find ClinicalTrials.gov study records, inspect exact NCT IDs and preserve source status, posted-results availability, processing dates and actual or estimated milestones.
Research NASA image records with AI and MCP
Find indexed space-image records, inspect exact NASA IDs and keep source dates, recorded credits and reuse uncertainty visible in a ToolCargo research workflow.
Check CVE aliases against CISA KEV with AI and MCP
Use ToolCargo OSV and CISA KEV to connect package advisories with recorded known exploitation, preserving dated source evidence and federal deadline context.
Find image candidates and inspect their credits with AI and MCP
Use ToolCargo’s Wikimedia Commons MCP to find image candidates, compare dimensions and inspect creator credits and recorded license terms before reuse.
How to run an SEO audit with an AI agent and MCP
Run a repeatable SEO audit with ToolCargo Site Audit: check served HTML, indexing signals and link previews, then verify fixes with your AI agent.
Find keyword opportunities with Google Search Console and MCP
Use Search Console query and page data in your AI agent to prioritize content updates, review CTR and find existing pages near the first page of search.
Compare npm and PyPI packages with an AI agent
Research JavaScript and Python dependencies through MCP using public npm and PyPI metadata, source links and explicit comparison limits.
Monitor SEO changes after launch with an AI agent and MCP
Set up scheduled SEO monitoring with ToolCargo Page Watch: track noindex, canonical and HTTP changes, review alerts, and verify fixes after a website launch.
Find research papers and verify DOI metadata with an AI agent
Build a source-linked research shortlist with ToolCargo Crossref and Wikipedia MCP tools. Check DOI records, preserve missing fields and separate metadata from paper claims.
Check npm and PyPI package vulnerabilities with an AI agent
Use ToolCargo OSV MCP to check exact package versions, inspect source-linked advisories and record the evidence needed for dependency review.
Find life-sciences publications with Europe PMC and MCP
Search Europe PMC publication metadata from your AI agent, inspect source records and build a cited reading list with dates and publication types.
Find research datasets with an AI agent and DataCite MCP
Find DataCite dataset DOI records with your AI agent, inspect deposited rights and build a source-linked shortlist before choosing data to reuse.
Review Rust crate versions and dependencies with MCP
Use crates.io release metadata and OSV advisory records in your AI agent to review an exact Rust package version before changing dependencies.
Research species and biodiversity records with MCP
Use GBIF MCP tools to find scientific names, inspect taxonomy and summarize indexed occurrence records with explicit country and year filters.
Compare public AI model and dataset metadata with MCP
Use Hugging Face Hub MCP tools to discover public models and datasets, review repository metadata and preserve source licenses, task tags and access limits.
Find books and compare editions with MCP
Use Open Library MCP tools to find a work, inspect its catalogue record and compare edition dates, publishers, languages and identifiers.
Resolve entities and review Wikidata statements with MCP
Use Wikidata MCP tools to resolve English or Arabic names, inspect an exact entity and review a specific property with ranks, qualifiers and references.
Review US weather forecasts with an AI agent and MCP
Use NWS MCP tools to resolve a US forecast grid and compare dated period and hourly forecasts with source units and missing values.
Compare country indicators with AI and World Bank MCP
Compare countries using ToolCargo World Bank MCP: choose one annual indicator and year, keep units and missing values explicit, and verify the source.
01 / US weather planning
Keep the forecast attached to the plan.
You are comparing weather periods for an outdoor event or a US trip.
A prompt to try after connecting
“Resolve these public US coordinates with NWS, inspect the period forecast and the relevant hourly periods in one unit system. Keep source update time, forecast start/end times, precipitation probabilities, null values and source links in the summary. Recheck the current forecast before making plans.”
How the workflow fits together
- 1Choose coordinates you intend to share with NWS and resolve the current forecast office and grid.
- 2Read the grid forecast, then inspect explicit hourly periods for your planned time window.
- 3Keep issue and valid times with the result, distinguishing forecasts from measured conditions.
What you get
A source-linked weather planning brief with dated periods and consistent units.
Know the limits
NWS-covered US locations only. Grid forecasts cover an area and can change; no measured conditions, alerts or emergency service. Missing probabilities are unknown, not zero. Review current official forecasts and guidance before decisions.
02 / Research
Build a brief with sources you can follow.
You need context on a topic before writing, planning or making a decision.
A prompt to try after connecting
“Find background on this topic in Wikipedia, look up related publications in Crossref and find relevant Hacker News discussions. Keep source links and distinguish published metadata from community opinions.”
How the workflow fits together
- 1Search English Wikipedia and read a bounded article extract with its revision details. Look up related scholarly publication metadata and DOIs with Crossref.
- 2Find relevant Hacker News stories and inspect a bounded sample of comments.
- 3Ask your agent to draft a brief with links and clearly identified gaps.
What you get
A source-linked starting point for further research.
Know the limits
Wikipedia can be incomplete or outdated. Crossref returns publication metadata, not full papers or peer-review judgments. Hacker News comments are opinions, and the connector returns a sample of a discussion.
03 / Entity research
Resolve the name. Review the statements.
You need to distinguish similarly named places, works or organizations and inspect recorded properties.
A prompt to try after connecting
“Search Wikidata for this name in English or Arabic. Review candidate labels and descriptions, inspect the selected exact Q ID, then retrieve one specific P-property’s statements. Keep typed values, ranks, qualifiers, references and missing values visible. Do not treat the result as independently verified fact.”
How the workflow fits together
- 1Search a name and compare candidates; shared labels do not prove the same identity.
- 2Inspect the exact entity and selected language fields, preserving source revision and missing metadata.
- 3Retrieve a chosen property and review ranks, qualifiers and source references. Request another explicit page only when needed.
What you get
A traceable entity and property review with source context and limits.
Know the limits
Community-edited statements can be incomplete, outdated or disputed. Preferred rank is a Wikidata editorial status, not independent verification. Missing, unknown and no-value snaks remain distinct; bounded output may omit some source context.
04 / Book research
Find the work. Inspect the edition.
You are researching books or choosing a particular edition for reading or citation.
A prompt to try after connecting
“Search Open Library for this title and author. Inspect the exact work record, list one page of editions, then inspect an edition linked to that work. Compare catalogue publication dates, publishers, languages and identifiers, and keep missing metadata explicit.”
How the workflow fits together
- 1Search by title and author, then inspect candidate work IDs rather than assuming a title match is unique.
- 2List editions of the selected work and inspect a specific edition ID. Keep work-level and edition-level dates separate.
- 3Build a source-linked shortlist and review the original records before choosing an edition.
What you get
Book and edition records with traceable identifiers and catalogue limits.
Know the limits
Human-directed discovery only, not bulk harvesting. Catalogue metadata can be incomplete or revised. No full text, book downloads, availability guarantees or reading rights are supplied. Results may use a five-minute metadata cache.
05 / AI model research
Inspect a model before choosing it.
You are shortlisting a public AI model or dataset for a project.
A prompt to try after connecting
“Search public Hugging Face model and dataset repository names or IDs. Inspect candidate repositories, capture commit SHAs and compare declared task, library, language, license and access metadata for my project. Keep missing fields explicit and do not infer quality or compatibility from popularity.”
How the workflow fits together
- 1Search by repository name or ID, then assess the returned task metadata. Search does not match task descriptions or filter by task, language or license.
- 2Read repository metadata and record the returned commit. Review declared licenses and access flags at their source.
- 3Explain what still needs task-specific evaluation and source review before downloading or running anything.
What you get
A source-linked shortlist of repositories and the checks that remain.
Know the limits
Metadata and popularity do not establish performance, safety, suitability or licensing permission. Public metadata can describe gated repositories. No files, dataset rows, model execution or inference are fetched.
06 / Biodiversity research
Ground a species question in source records.
You are researching taxonomy or reported biodiversity observations.
A prompt to try after connecting
“Find this scientific name in GBIF, inspect the taxonomy and confirm the backbone occurrence taxon key. Summarize indexed occurrence records for my country and year, keeping source limits explicit.”
How the workflow fits together
- 1Search scientific names and inspect checklist provenance, taxonomic status and synonyms.
- 2Review exact taxonomy records and select the occurrence taxon key instead of assuming a search match.
- 3Summarize occurrence records using that key and explicit filters. Keep counts, partial facets and lookup links.
What you get
Source-linked taxonomy and an overview of indexed occurrence records.
Know the limits
Occurrence counts reflect reported records, not populations, current presence or a complete survey. Zero records does not establish absence. Review original datasets and citation requirements before reuse.
07 / Rust dependencies
Inspect a crate before you build with it.
You are reviewing a Rust dependency or planning a version change.
A prompt to try after connecting
“Find this Rust crate, inspect the exact version we plan to use, list its declared dependencies and check that same package/version in OSV with ecosystem crates.io. Keep yanked releases, prereleases, missing fields and advisory coverage explicit.”
How the workflow fits together
- 1Search crates.io and confirm the exact crate name.
- 2Inspect release metadata and declared dependencies using an explicit version. Review license declarations, Rust-version requirements and yanked status.
- 3Query OSV for the same crates.io package/version and investigate any advisory links.
What you get
A source-linked starting point for dependency review.
Know the limits
Declared dependencies are not a resolved lockfile or transitive dependency tree. Metadata and advisory absence do not establish safety, compatibility or license suitability. No installation or code execution occurs.
08 / Life-sciences literature
Find the papers behind a research question.
You are preparing a literature starting point for biology or life-sciences research.
A prompt to try after connecting
“Use Europe PMC to find publications on this topic within my year range. Inspect the source records, keep publication types and source links, and flag preprints or notices for review.”
How the workflow fits together
- 1Search publication metadata with a topic and optional year range.
- 2Inspect individual records by source and ID; retain publication types, dates and available open-access flags.
- 3Read the original papers and related notices before evaluating the evidence.
What you get
A bounded, source-linked list of publication records for further reading.
Know the limits
Publication metadata is not a full paper, evidence-quality assessment or medical advice. Open access does not by itself establish a reuse license. Publication types and notices may be incomplete.
09 / Datasets & institutions
Find a dataset you can investigate.
You need source data for a research project, brief or analysis, and want to identify the institutions named in its metadata.
A prompt to try after connecting
“Search DataCite for datasets about this topic, inspect relevant DOI records and keep creators, publication years, deposited rights and related identifiers visible. Look up a named institution in ROR and confirm its identity from the returned records. Link to sources and flag uncertain matches.”
How the workflow fits together
- 1Search a topic in DataCite with optional publication-year bounds, then inspect a dataset by its returned DOI.
- 2Read the deposited creators, publisher, rights and related identifiers. Follow the source to review actual files, access conditions and reuse terms.
- 3Search an institution name in ROR and inspect its ID, status, location and relationships. Confirm a match rather than assuming that similar names mean the same organization.
What you get
Source-linked dataset metadata and research-organization records for a documented starting point.
Know the limits
These tools do not download datasets, inspect data quality or grant reuse rights. DataCite and ROR are separate registries; a ROR match does not prove dataset authorship, ownership or accreditation. Deposited metadata can be incomplete or outdated.
010 / Development
Choose a dependency with more context.
You are comparing libraries for your next feature or an upgrade.
A prompt to try after connecting
“Compare these packages in their registry: npm for JavaScript or PyPI for Python. Show current versions, license metadata and dependencies, and link to the package sources. Check the exact versions in OSV for known vulnerability records.”
How the workflow fits together
- 1Inspect a public npm or PyPI package’s metadata and release information.
- 2Compare a small set of packages using the same fields, then query their exact versions in OSV and inspect any advisory records.
- 3Review the documentation and repository before choosing what to install.
What you get
A structured comparison grounded in registry metadata and source-linked known advisories.
Know the limits
Package metadata and popularity do not prove security, quality or compatibility. No OSV matches does not prove a version is safe; coverage, withdrawn records and omitted results matter. These tools do not install or execute packages.
011 / Earthquake research
Check the event behind the headline.
You are researching a reported earthquake or comparing dated catalogue records.
A prompt to try after connecting
“Search the USGS catalogue for earthquakes in this date range, filter by magnitude and location, then inspect a relevant event. Keep event time, last update, depth units and source links clear.”
How the workflow fits together
- 1Search a bounded date range with an optional location and minimum magnitude.
- 2Page through results if needed, then inspect an event by its returned USGS ID.
- 3Build a dated factual summary that keeps unknown values and revisions visible.
What you get
Event records with magnitude, coordinates, depth, UTC timestamps and official source links.
Know the limits
Catalogue entries can be revised, delayed or incomplete. This is an on-demand research tool, not forecasting, an alert service or emergency guidance.
012 / Website launches
Hand over a site with the checks attached.
You are about to launch a new site or a redesigned client page.
A prompt to try after connecting
“Audit the home, pricing and contact pages. Prioritize indexing blockers, broken links and social preview issues, then give me a fix list.”
How the workflow fits together
- 1Audit key pages individually or use audit_site for a capped selection of pages.
- 2Review high-impact findings and fix staging noindex tags, canonicals, titles and broken previews.
- 3Re-run the saved audits and compare the results for your handover.
What you get
Dated findings with evidence, fixes and a before/after comparison.
Know the limits
Site Audit reads served HTML and does not run JavaScript. A capped multi-page audit is not a complete site crawl.
013 / Ongoing care
Keep an eye on what changed after launch.
You maintain client sites and want a clear record of regressions and recoveries.
A prompt to try after connecting
“Watch our important pages weekly for indexing changes. Show the change history when I ask, and help me review our uptime checks.”
How the workflow fits together
- 1Add important public pages to Page Watch and confirm your email for alerts.
- 2Choose a supported daily or weekly schedule within your plan; use Uptime & SSL Monitor for HTTP and certificate checks.
- 3Review watch history and saved audit comparisons during your regular client check-in.
What you get
Scheduled checks, change history and alerts for supported transitions.
Know the limits
Page Watch runs from an hourly job when due; it is not continuous monitoring. Checks consume the shared quota. Saved audit re-runs can also be started manually.
014 / Campaigns & content
Give a shared link the right first impression.
A page’s social preview shows a missing image or the wrong title.
A prompt to try after connecting
“Check the social preview for this page. Show which tags or image details need changing and explain the evidence.”
How the workflow fits together
- 1Run check_social_preview on the public URL.
- 2Inspect the returned tags and image status, type, size and dimensions.
- 3Fix the markup and use the platform’s own refresh tool to update its cache.
What you get
Specific preview findings you can pass to a developer.
Know the limits
Social platforms apply their own rules and caching. A passing tag check cannot guarantee how a platform renders the preview.
015 / Business reviews
Bring the numbers into the conversation.
You want to review performance with your agent, then share a useful summary.
A prompt to try after connecting
“Review our connected analytics and search performance. Separate the reporting period from today, then summarize what needs a closer look.”
How the workflow fits together
- 1Connect your own supported provider account with the documented read permissions.
- 2Request the relevant metrics and keep the time range and source clear.
- 3Pair traffic context with website findings, or export a saved audit as Markdown or a printable report.
What you get
Provider data and page findings your agent can help explain.
Know the limits
Provider permissions, plans, data delays and row limits apply. Google connections currently require invited test accounts. An audit score is not a ranking or a traffic forecast.
Your next task
Find the tools that fit.
Explore MCP servers without an account. To use ToolCargo’s hosted connectors, create a free account and follow the setup for your AI client. You can also run Site Audit in your browser.