Panda Federation Tools

Panda Federation Tools

An open-source MCP server and tool index for 18 research and public-data tools on Apify. Check citations for fake or retracted papers, find real literature, build grant, contract and lead lists, and audit domains and websites, from Claude, Cursor, VS Code or any MCP client, or straight from the Apify API.

• Official, free data sources only: Crossref, OpenAlex, PubMed, Unpaywall, Grants.gov, NIH, NSF, USAspending, NPI Registry, FMCSA, public DNS, Apple Podcasts, OpenStreetMap, Wikimedia, public job-board feeds and companies' own websites. No logins, no paywalls, no LinkedIn scraping.
• Pay per result, capped: a few tenths of a cent per result; runs that find nothing are not charged per result; every MCP call has a hard USD spending cap.
• Agent-friendly output: JSON with a one-line summary or agentSummary per row and OK / CHECK / FIX status where it applies.
• Your token stays yours: the server runs on your machine and sends your APIFY_TOKEN only to api.apify.com, in a header. It is never logged or returned.

Contents

• Quick start
• Client setup (Claude Desktop, Claude Code, Cursor, VS Code, others)
• Tools
• Pricing
• Configuration
• Security and permissions
• Hosted options (no install)
• For AI agents and LLMs
• Development

Quick start

1. Get a free Apify account and copy your API token from Console → Settings → API & Integrations. The free plan includes $5 of monthly credit.
2. Install uv (one command; it brings its own Python).
3. Check that the server starts:

uvx --from git+https://github.com/pandadad0214/[email protected] panda-federation-tools --version

Then add it to your MCP client below. The server speaks MCP over stdio.

Client setup

Replace apify_api_YOUR_TOKEN with your own token. Do not commit config files that contain it.

Claude Desktop

Edit claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "panda-federation-tools": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/pandadad0214/[email protected]", "panda-federation-tools"],
      "env": { "APIFY_TOKEN": "apify_api_YOUR_TOKEN" }
    }
  }
}

Claude Code

claude mcp add panda-federation-tools --env APIFY_TOKEN=apify_api_YOUR_TOKEN -- \
  uvx --from git+https://github.com/pandadad0214/[email protected] panda-federation-tools

Cursor

Add to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):

{
  "mcpServers": {
    "panda-federation-tools": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/pandadad0214/[email protected]", "panda-federation-tools"],
      "env": { "APIFY_TOKEN": "apify_api_YOUR_TOKEN" }
    }
  }
}

VS Code (GitHub Copilot agent mode)

Add to .vscode/mcp.json. VS Code prompts for the token and stores it securely:

{
  "inputs": [
    { "type": "promptString", "id": "apify-token", "description": "Apify API token", "password": true }
  ],
  "servers": {
    "panda-federation-tools": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--from", "git+https://github.com/pandadad0214/[email protected]", "panda-federation-tools"],
      "env": { "APIFY_TOKEN": "${input:apify-token}" }
    }
  }
}

Windsurf, Cline, Zed, Goose and other MCP clients

Use the same stdio command in that client's MCP settings:

• Command: uvx
• Args: --from git+https://github.com/pandadad0214/[email protected] panda-federation-tools
• Env: APIFY_TOKEN=apify_api_YOUR_TOKEN
• Cline: also set "timeout": 60 (seconds). Starting with uvx from GitHub can take longer than Cline's 10-second default. Step-by-step setup is in llms-install.md.

Prefer pip? pip install "git+https://github.com/pandadad0214/[email protected]", then run panda-federation-tools as the command.

Prefer Docker? Build the image from this repo and run it over stdio:

docker build -t panda-federation-tools https://github.com/pandadad0214/panda-federation-tools.git#v1.1.1
docker run -i --rm -e APIFY_TOKEN panda-federation-tools

Tools

Each MCP tool runs one public Apify Actor and returns its results. Input fields, limits and defaults come from the Actor's own input schema and are validated before anything runs, so a bad input never costs money.

Every tool also accepts three optional run controls: run_max_items (results to return, default 50), run_max_charge_usd (spending cap for the call, default $0.50) and run_timeout_secs (default 120, max 300). While a run is in progress the server sends MCP progress notifications every few seconds, so clients that support them do not time out. Most tools answer in 5–15 seconds for a few inputs; local_business_leads and big crawls can take a minute or more, so give them run_timeout_secs: 300.

Also on Apify: the hosted Academic Research MCP Server ($0.01–$0.02 per call) and Website Change Monitor (publishing soon; it joins this server once it is public).

Example prompts

• "Check these 12 references for fake or retracted papers and tell me which to fix."
• "Find 20 recent peer-reviewed papers on GLP-1 and heart failure with free PDFs, then format them in APA."
• "Which of these 30 SaaS companies are hiring sales leaders right now? Then find their VP-level contacts."
• "Audit the email security of acme.com and its three competitors, and list the fixes."
• "List dentists in Phoenix, AZ with a website and phone number."

Typical workflows

• Writing a paper: literature_search → citation_formatter → reference_checker before you submit.
• Systematic review: pubmed_evidence_table (evidence table + PRISMA counts) → reference_checker.
• Funding: grants_finder → literature_search on the topic.
• Federal contracting: federal_contract_finder (recompete radar + top incumbents) → grants_finder.
• Account prioritization: company_hiring_signals → company_website_enricher → email_domain_security_checker.
• Outbound to decision makers: company_hiring_signals → company_team_finder (names, titles, email pattern).
• Local prospecting: local_business_leads → ai_search_seo_audit on their websites.
• Trucking and logistics sales: trucking_company_leads → company_website_enricher.
• Market research: topic_trends_tracker → literature_search or podcast_leads_finder on rising topics.

Pricing

• You pay Apify per result, at the prices above, from your own Apify account. Apify Bronze, Silver and Gold plans get 10–30% off.
• Empty runs cost nothing per result. Apify adds a tiny platform start fee (about $0.00005 per run).
• run_max_items (default 50) also sets the Actor's own result limit on 15 of the 18 tools, so you only pay for the results you asked for. On grants_finder, pdf_table_extractor and company_team_finder (whose limits count sources, pages or companies), set those fields directly; the charge cap still applies.
• Every MCP call is capped by run_max_charge_usd (default $0.50, set your own default with PANDA_MAX_CHARGE_USD). Apify stops the run when the cap is reached.
• The MCP server itself is free and open source (MIT).

Configuration

Security and permissions

• Runs locally over stdio. It opens no ports and calls only https://api.apify.com.
• Token handling: read from the environment, sent only as an Authorization: Bearer header (never in a URL), redacted from every error message, and never logged or returned to the model.
• Least privilege: the token needs permission to run Actors. Apify supports scoped API tokens; a token limited to running these Actors is enough.
• Read-only tools: every tool only reads public data; none writes to your systems. Tools are marked readOnlyHint: true.
• Validated inputs: every call is checked against the Actor's JSON Schema (types, ranges, enums, no unknown fields) before it runs. Secret Actor inputs (such as optional third-party API keys) are deliberately not exposed to the model.
• Bounded cost and time: per-call spending cap, item cap and timeout. The server retries once only when Apify never started a run (rate limit or failed connection), so a retry can never charge you twice.
• Context-safe output: large results are trimmed to whole items (still valid JSON) for the model, with every item kept in structuredContent.
• Supply chain: pinned dependencies with a committed uv.lock, pip-audit and CodeQL in CI, Dependabot updates, and GitHub Actions pinned to commit SHAs.
• Found a vulnerability? See SECURITY.md.

Hosted options (no install)

• Academic Research MCP Server (Streamable HTTP): https://pandadad0214--panda-research-mcp.apify.actor/mcp with header Authorization: Bearer <your Apify token>. Four research tools (check_references, format_citations, search_literature, pubmed_evidence_table). This is the server listed in the official MCP Registry as io.github.pandadad0214/panda-research-mcp (server.json).
• Apify's MCP server: https://mcp.apify.com can call any of these Actors as pandadad0214/<slug>.
• Plain HTTP API:

curl -X POST "https://api.apify.com/v2/acts/pandadad0214~reference-checker/run-sync-get-dataset-items" \
  -H "Authorization: Bearer $APIFY_TOKEN" -H "Content-Type: application/json" \
  -d '{"dois": ["10.1038/nature14539"]}'

For AI agents and LLMs

• llms.txt: short index of every tool and when to use it (llms.txt convention).
• llms-full.txt: each tool's inputs, outputs and prices in one file.
• llms-install.md: step-by-step setup for AI coding agents such as Cline.
• Every result has a one-line summary field (summary on the citation, literature, PubMed, PDF, NPI and expert tools; agentSummary on the others), and every run saves a SUMMARY record with a takeaway and the suggested next tool.
• Start small (one or two inputs), read the summary line, then widen.

Development

git clone https://github.com/pandadad0214/panda-federation-tools && cd panda-federation-tools
uv sync                 # installs pinned dependencies from uv.lock
uv run pytest           # offline tests, no token or network needed
uv run ruff check . && uv run mypy
python scripts/build_catalog.py   # refresh tool schemas and prices from the Apify Store

Layout: src/panda_federation_tools/ (server, Apify client, bundled catalog.json), tests/, scripts/build_catalog.py, Dockerfile. See CONTRIBUTING.md and CHANGELOG.md.

Hire a human (Fiverr)

Prefer a person to run the tools and hand you finished files? Panda Federation also sells these as Fiverr gigs (seller pandafederation). Professional use only, not for coursework.

Issues and requests

Open an issue here or on any tool's Apify Issues tab. We answer within a day.

Built by Panda Federation with AI assistance. MIT licensed.