This MCP server gives Claude six tools to estimate at-berth CO₂ for tankers and query Brazilian port statistics (247 ANTAQ installations, 2010–2026) with sources and method caveats attached.
estimate_berth_co2 — at-berth auxiliary-engine CO₂ for one liquid-bulk call (DWT, berth hours, fuel, optional SFOC), i.e. the CO₂ shore power would avoid; boilers excluded.
estimate_calls_co2 — the same calculation summed across many calls, with totals by size band.
find_installation — resolve a name, port complex or ANTAQ code (e.g. "Ponta da Madeira" → BRMA002) to codes usable in other tools, returning candidates when ambiguous.
port_profile — year-by-year cargo (tonnes), navigation-type split, berthings and mean port times (waiting, operating, at berth, total stay) for one installation over a chosen period.
compare_ports — rank multiple installations in one year on tonnes, berthings, waiting hours, hours at berth or total stay.
explain_method — surface the formula, parameters and sources behind berth CO₂, port statistics or port times.
Works from a PORT_DATA_DIR folder, a one-command Zenodo download of the open ANTAQ dataset, or bundled synthetic data (flagged as invented), and instructs Claude to cite sources and flag partial years like 2026.
Fetches the open dataset 'Brazil Port Data: consolidated ANTAQ port statistics, 2010-2026' from the Zenodo API, providing the MCP server with real port data for its tools.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@port-emissions-mcp Rank Itaqui, Santos and Paranaguá by average waiting time in 2025."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP server that lets Claude estimate the at-berth CO₂ of tankers and answer questions about Brazilian ports (247 ANTAQ installations, 2010 to 2026), with the source and the method behind every number
The Model Context Protocol (MCP) lets an AI assistant call external tools. This server gives Claude six tools that join two pieces of my work: the maritime-co2 library (at-berth CO₂ of liquid-bulk vessels, IMO Fourth GHG Study 2020 method) and the Brazil Port Data tables (ANTAQ statistics aggregated with the same filter as the agency's public panel).
With it, Claude can answer questions such as:
"How long did ships wait to berth at Itaqui, Paranaguá, Santos and Ponta da Madeira in 2025? Rank them."
"Give me the cargo profile of Itaqui since 2019, split by deep sea and cabotage."
"A 45,000 DWT tanker stayed 67 hours at berth. How much CO₂ would shore power have avoided?"
"Here are 40 calls with DWT and hours at berth. What is the total OPS-avoidable CO₂ by size band?"
Tool | What it does |
| Searches installations by name, port complex or ANTAQ code ("Ponta da Madeira" → |
| Year-by-year cargo (tonnes), navigation mix, berthings and port times for one installation |
| Ranks several installations on tonnes, berthings, waiting time, hours at berth or total stay |
| At-berth CO₂ of one tanker call (auxiliary engine, the load shore power replaces) |
| The same for a list of calls, with totals by size band |
| Formula, parameters and sources behind the numbers |
Design choices that matter when an LLM is the user:
Every answer carries its source, and the server instructions tell Claude to cite it and to flag partial years (2026 rows cover only part of the year).
Names resolve to codes safely. An exact name wins ("Paranaguá" is the public port, not the Cattalini terminal); an ambiguous name returns the candidate codes so Claude can ask or choose, instead of guessing.
Errors are written for the model. "No installation matches 'xyz'. Use find_installation first." lets Claude correct itself in the next step.
Caveats travel with the CO₂ numbers. Auxiliary power and load factor are illustrative defaults by DWT band, and the tool says so.
Related MCP server: Datalastic MCP Server
The server looks for the port tables in this order:
The folder in PORT_DATA_DIR, if set (for your own copy or a newer extraction).
The open dataset on Zenodo, after one command:
port-emissions-mcp downloadThis fetches Brazil Port Data: consolidated ANTAQ port statistics, 2010-2026 (doi:10.5281/zenodo.23158267, ODbL 1.0) into ~/.cache/port-emissions-mcp.
A small synthetic dataset bundled in src/port_emissions_mcp/example_data/ (three fictitious installations), so the tests run offline. Every answer based on it carries a warning that the numbers are invented.
File | Grain |
| year × installation, tonnes |
| year × installation, tonnes by navigation type |
| year × installation, berthings |
| year × installation, mean hours waiting, operating, at berth, total stay |
| installation names (linked to ANTAQ codes by tonnage in 2024 and 2025, one-to-one matches only) |
Requires Python 3.10+ and uv (or pip).
git clone https://github.com/darlianecunha/port-emissions-mcp
cd port-emissions-mcp
uv venv && uv pip install -e ".[test]"
uv run pytest # 10 tests, offline, synthetic data
uv run port-emissions-mcp download # real data from Zenodoclaude mcp add --transport stdio --env PORT_DATA_DIR=/path/to/dados-consolidados \
port-emissions -- /path/to/port-emissions-mcp/.venv/bin/port-emissions-mcpLeave out --env ... if you ran port-emissions-mcp download. Check with claude mcp list.
Add to claude_desktop_config.json (Settings → Developer → Edit Config) and restart the app:
{
"mcpServers": {
"port-emissions": {
"command": "/path/to/port-emissions-mcp/.venv/bin/port-emissions-mcp",
"env": {}
}
}
}Path | Content |
| The six MCP tools and the server instructions |
| Loading, name-to-code linking, search and resolution |
| Fetches the open dataset from the Zenodo API (standard library only) |
| Synthetic data (invented numbers) |
| pytest suite: tools, name resolution, MCP exposure, and a download from a local mock of the Zenodo API |
| Prompts used to test the server with Claude |
At-berth CO₂ (via maritime-co2, doi:10.5281/zenodo.20708090): CO₂ = aux_kW × load_factor × hours × SFOC / 10⁶ × Cf, auxiliary engine only (boilers excluded because they keep running for cargo heating). Cf from the IMO Fourth GHG Study 2020: MDO/MGO 3.206, HFO 3.114 t CO₂/t fuel; SFOC 215 g/kWh. Scope: liquid-bulk (tanker) vessels.
Data licence. The port tables are ODbL 1.0, derived from ANTAQ open data; cite the Zenodo record. The code is MIT.
Cargo follows ANTAQ's definition of port movement (authorised operations that count as movement), so national totals match the agency's panel.
Berthings count distinct records with authorised cargo operations, so they are lower than raw ANTAQ berthing counts.
Port times are annual means per installation.
Limitations. Size-band parameters are illustrative, not vessel-specific. 2026 is partial. 71 of 247 codes have no linked name and are labelled with their port complex.
maritimeco2: the CO₂ library this server calls
decarbport.com: interactive calculator using the same method
brazilportdata.com: public panels built from the same ANTAQ tables
MCP_DadosPublicosANTAQ: a broad MCP server over many ANTAQ and trade archives. This project is narrower by design: curated annual series that match ANTAQ's official totals, plus at-berth CO₂ and method caveats in every answer
Cunha, D. R. (2026). port-emissions-mcp: an MCP server for at-berth CO₂ and Brazilian port statistics (Version 0.1.0) [Software]. https://github.com/darlianecunha/port-emissions-mcp
Darliane Ribeiro Cunha, PhD. ribeirocunha.com · ORCID 0000-0003-2548-1237
MIT.
Compare several installations on one indicator in one year, ranked high to low.
Args: installations: ANTAQ codes or unambiguous names. indicator: tonnes, berthings, waiting_hours, hours_at_berth or total_stay_hours. year: reference year.
Inputs: indicator (required), installations (required), year.
Estimate the at-berth CO2 of ONE liquid-bulk (tanker) call from the auxiliary engine, i.e. the CO2 that Onshore Power Supply (shore power) would avoid. Boilers are excluded.
Args: dwt: deadweight tonnage of the vessel. berth_hours: hours at berth (from berthing to unberthing). fuel: fuel burnt by the auxiliary engine. sfoc_g_per_kwh: specific fuel oil consumption, default 215 g/kWh.
Inputs: dwt (required), berth_hours (required), fuel, sfoc_g_per_kwh.
Estimate at-berth (OPS-avoidable) CO2 for MANY liquid-bulk calls and sum it.
Args: calls: list of objects with keys "dwt" and "berth_hours" (optional "call_id"). fuel: fuel burnt by the auxiliary engine.
Inputs: calls (required), fuel.
Explain the method, parameters and sources behind a tool's numbers.
Inputs: topic (required). Returns: result
Search Brazilian port installations (public ports and private terminals) by name, port complex or ANTAQ code, e.g. "Itaqui", "Ponta da Madeira", "Paranagua", "BRSSZ". Returns ANTAQ codes to use in the other tools.
Inputs: query (required).
Year-by-year profile of one installation: cargo handled (tonnes), split by navigation type, number of cargo berthings and mean port times (waiting, operation, at berth).
Args: installation: ANTAQ code (e.g. BRIQI) or an unambiguous name. start_year, end_year: period, inclusive (data from 2010).
Inputs: installation (required), end_year, start_year.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 6 tools
The two estimator tools are distinguished by scale (ONE call vs MANY calls), and port_profile (single installation over time) vs compare_ports (several installations in one year) are clearly differentiated. There is minor overlap in purpose between the two CO2 estimators, but descriptions make the intent clear.
Most tools follow a verb_noun pattern (estimate_berth_co2, estimate_calls_co2, find_installation, compare_ports, explain_method). port_profile deviates from this pattern by using noun_noun, a minor inconsistency.
Six tools is well-scoped for a niche port-emissions analysis domain. Each tool—single estimation, batch estimation, lookup, profile, comparison, and method explanation—earns its place without redundancy.
The surface covers the core lifecycle: finding installations, profiling, comparing, estimating emissions, and explaining methods. Minor gaps exist, such as no direct tool to enumerate all installations or aggregate multi-year comparisons, but agents can work around these.
IBGE: geography, census, economy and health from the official APIs, with provenance. 23 tools.
AI-powered commodity price forecasts and Brazil trade intelligence via MCP. Access Brazilian exports and imports from MDIC/ComexStat since 2000, by HS4, HS6 and NCM code, partner country, USD FOB and kilograms — plus crop production, supply & demand, climate data and production forecasts for major commodities including soybeans, coffee, corn, beef and cocoa. Compare trade flows, markets, countries and time periods with like-for-like windows. Every answer identifies the measurement period and data source. 15 read-only tools. Metered in credits. Connect directly via Streamable HTTP — no installation required: https://mcp.kyrodata.com/mcp Authentication: Bearer key or OAuth 2.1 (PKCE). Official MCP registry: com.kyrodata/kyrodata
JusPronto is case-management software for Brazilian law firms. Its remote MCP server (OAuth 2.1 with PKCE, per-permission scopes, audit log) lets Claude and ChatGPT read the firm's own cases, deadlines, agenda, recent court movements and case files with the page cited. Write actions require the lawyer's approval. Requires a JusPronto account.
Read-only Amazon profit, fee, ACoS, and geo-holdout MCP tools for Claude, ChatGPT, and agents.
