Live geospatial intelligence as MCP tools, bundled with an analyst skill, three slash commands, and a watch-officer agent. Sweep the planet for a few hundred tokens; drill on demand.
- 41 MCP tools over live ADS-B (aircraft), AIS (vessels), a GPS-jamming layer, Sentinel-1 SAR dark vessels, geocoded events, and a cross-domain incident-fusion engine.
- Context-optimised — every heavy tool takes
detail='short'(a token-frugal digest, the default) ordetail='long'(the full bundle). See the skill. - Skill
osint-intelteaches the brief-first → drill-second workflow, withreference/tools.mdandreference/workflows.mdloaded on demand. - Commands
/osint-brief,/osint-watch,/osint-jamming. - Agent
osint-watch-officerfor standing surveillance.
The tools query a local OSINT backend, so start it first (from the repo root):
bash scripts/run-api.sh # backend on :8000Full plugin (MCP + skill + commands + agent):
/plugin marketplace add /path/to/OSINT
/plugin install osint-geoint@osint-velocity
When prompted, set repo_dir to your cloned OSINT repo (the folder containing
apps/api) and python to that repo's venv interpreter —
apps/api/.venv/bin/python (macOS/Linux) or apps\api\.venv\Scripts\python.exe
(Windows). api_base defaults to http://127.0.0.1:8000. The plugin launches
that Python directly (python -m app.mcp_server), so the same manifest works on
Windows, macOS, and Linux.
MCP only (no skill/commands), or to script it — the installer prints the exact
commands for your OS and, with the run flag, registers the server via
claude mcp add:
# Linux / macOS (or double-click install.command on macOS)
bash plugin/osint-geoint/install.sh # print commands
bash plugin/osint-geoint/install.sh -y # also register the MCP server# Windows PowerShell (or double-click install.cmd)
plugin\osint-geoint\install.ps1 # print commands
plugin\osint-geoint\install.ps1 -Run # also register the MCP serverHosted (no local backend; needs a Velocity token):
claude mcp add --transport http osint-geoint \
https://projectvelocity.org/mcp \
--header "Authorization: Bearer $VELOCITY_TOKEN"get_situation() to orient → intel_brief() for cited cross-domain incidents →
focus_area(lat, lon) to load a region PRIMARY → drill with query_vessels /
gps_jamming / detect_deception → deep_analyze() for an off-context judgement.
Full architecture and the /api/intel/* HTTP surface: docs/mcp-server.md.