atlaspi-mcp

atlaspi-mcp

mcp-name: io.github.Soil911/atlaspi

The first MCP server for historical geography. Wraps AtlasPI, a structured historical-geographic database (860+ geopolitical entities, 430+ events, 110+ cities across 65,000 years), so any MCP-compatible client — Claude Desktop, Claude Code, Cursor, Zed, etc. — can answer questions like "What empire ruled Egypt in 300 BC?" with sourced, structured data instead of hallucinating.

What is AtlasPI?

AtlasPI is a structured historical-geographic database, designed to be consumed by AI agents. It provides coordinates, GeoJSON boundaries and historical metadata of geopolitical entities in any era. Live API and docs: https://atlaspi.it.

This package (atlaspi-mcp) is the official Model Context Protocol bridge.

Install

pip install atlaspi-mcp

Requires Python 3.10+.

Configuration

Claude Desktop

Edit your claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json, Windows: %APPDATA%\Claude\claude_desktop_config.json):

{
  "mcpServers": {
    "atlaspi": {
      "command": "python",
      "args": ["-m", "atlaspi_mcp"]
    }
  }
}

Restart Claude Desktop. The 25 AtlasPI tools will appear in the tools menu.

Claude Code

Add the server to your project or user MCP config:

claude mcp add atlaspi -- python -m atlaspi_mcp

Or edit ~/.claude/mcp.json (or the project-local .mcp.json) directly:

{
  "mcpServers": {
    "atlaspi": {
      "command": "python",
      "args": ["-m", "atlaspi_mcp"]
    }
  }
}

Pointing to a custom AtlasPI instance

By default the server talks to the public production API (https://atlaspi.it). To target a self-hosted or staging instance, set the ATLASPI_API_URL environment variable:

{
  "mcpServers": {
    "atlaspi": {
      "command": "python",
      "args": ["-m", "atlaspi_mcp"],
      "env": {
        "ATLASPI_API_URL": "https://your-atlaspi.example.com"
      }
    }
  }
}

Tools exposed

v0.1 — core entities

Tool	What it does
search_entities	Filter entities by name, year, type, continent, status
get_entity	Full detail (boundary, capital, sources) for an entity by ID
snapshot_at_year	All entities active in a given year (optionally by region/type)
nearby_entities	Entities near a (lat, lon) coordinate, optionally pinned to a year
compare_entities	Side-by-side comparison of two entities by ID
random_entity	Random entity, with optional filters (great for exploration)
get_evolution	Timeline of territorial changes for an entity
dataset_stats	Aggregate stats: totals per type, continent, status, year coverage

v0.2 — events, cities, routes, chains

Tool	What it does
search_events	Filter historical events (ETHICS-007: GENOCIDE, COLONIAL_VIOLENCE, no euphemisms)
get_event	Event detail with entity roles (MAIN_ACTOR, VICTIM, ...) and sources
events_for_entity	All events where a given entity appears, with optional role filter
search_cities	Filter cities by year, type (TRADE_HUB, CAPITAL, ...), entity, bbox
get_city	City detail with name_variants (ETHICS-009: colonial renames documented)
search_routes	Filter trade routes; involves_slavery surfaces slave-trade routes (ETHICS-010)
get_route	Route detail with waypoints, commodities, geometry
search_chains	Find dynasty/succession/colonial/ideological chains, filter by region/year/type
get_chain	Chain detail with ordered links + explicit transition_type per link (ETHICS-002)
entity_predecessors	Predecessors of an entity in chains (who came before, via which transition)
entity_successors	Successors of an entity in chains (who came after, via which transition)
what_changed_between	Diff two world snapshots: entities appeared / disappeared / persisted

v0.3 — unified timeline + fuzzy search

Tool	What it does
full_timeline_for_entity	Merged chronological stream of events + territory_changes + chain transitions for one entity
fuzzy_search	Cross-script approximate name search (difflib) — robust to spelling variants and transliteration
nearest_historical_city	Find historical cities nearest to (lat, lon) in a given year (haversine, composite client-side)

v0.4 — events on map + ephemeris

Tool	What it does
events_for_map	Lightweight event payload for map overlay: only events with coordinates, auto-expanding time window
on_this_day	Events that occurred on a specific MM-DD date across all years — great for "on this day" content

Example prompts

Try these in Claude Desktop / Claude Code once the server is configured:

• "What empire ruled Egypt in 300 BC?"
• "Compare the Roman Republic and the Byzantine Empire."
• "What was near coordinates 41.9, 12.5 in 100 AD?"
• "Show me the political snapshot of Europe in 1500."
• "Pick a random Asian khanate and tell me its history."
• "How did the Ottoman Empire's territory evolve over time?"

The agent will combine multiple tools (e.g. snapshot_at_year -> get_entity -> get_evolution) and answer with citations from the underlying dataset.

For AI agents

AtlasPI is built specifically for retrieval by LLM agents. Each record exposes:

• name_original + name_original_lang (always the local/historical name)
• name_variants[] (other languages, with sources)
• boundary_geojson (GeoJSON polygon, ready for spatial reasoning)
• confidence_score (0.0 -> 1.0 — surface uncertainty, do not fake certainty)
• sources[] (citable references)
• acquisition_method for territorial events (conquest, treaty, dynastic, ...)

Workflow tip: when the user asks an open-ended historical question, call dataset_stats once to ground the model on coverage, then chain search_entities -> get_entity -> get_evolution for a full answer.

Local testing

# install in editable mode with dev deps
pip install -e ".[dev]"

# run the unit tests
pytest

# (optional) run the live integration smoke test
ATLASPI_RUN_INTEGRATION=1 pytest -k integration

# launch the server manually (it speaks MCP over stdio)
python -m atlaspi_mcp

Publishing to PyPI

cd mcp-server

# Build
python -m build

# Upload to PyPI (requires API token)
python -m twine upload dist/atlaspi_mcp-0.4.0*

To set up PyPI credentials:

• Create an account on https://pypi.org
• Create an API token at https://pypi.org/manage/account/token/
• Use python -m twine upload --username __token__ --password pypi-xxx dist/*

Links

• AtlasPI live API: https://atlaspi.it
• AtlasPI source code: https://github.com/Soil911/AtlasPI
• Model Context Protocol: https://modelcontextprotocol.io

License

Apache-2.0 — see LICENSE.

The AtlasPI dataset itself follows an open-core model; check the main repo for dataset licenses and attribution requirements.