For agents
MCP & agent access
The Internet Compass is a static site: there is no running server to hold an MCP session. What's published instead is the same data surface described in MCP tool and resource shape, backed by plain GET endpoints that work with any HTTP client today.
Why there's no live MCP transport
MCP requires a stateful session over stdio or Streamable HTTP, which needs a server process. This site ships as static HTML and JSON, deployed to a CDN with no process behind it. Rather than claim a protocol endpoint that doesn't exist, /api/mcp publishes a descriptor: the same tools and resources an MCP server for this data would expose, mapped to the plain HTTP endpoints that already serve that data.
If you're building an MCP server against this dataset, the descriptor at /api/mcp is a reasonable starting map: one get_* tool per entity type, each resolving to a real, working URL.
Data surface
Entity types available
| Type | URL pattern | Records |
|---|---|---|
| Company | /companies/{slug} | 18 |
| Software | /software/{slug} | 48 |
| Category | /categories/{slug} | 14 |
| City | /cities/{slug} | 12 |
| Role | /salaries/{slug} | 14 |
| Term | /glossary/{slug} | 50 |
| Launch | /products/{slug} | 12 |
| Publication | /publications/{slug} | 10 |
| Article | /analysis/{slug} | 10 |
Endpoints
All agent-facing endpoints
- /llms.txttext/markdown
Curated index of the site for language models
- /llms-full.txttext/markdown
Expanded index with inlined reference content
- /ai.txttext/plain
AI usage, attribution and citation policy
- /api/snapshotapplication/json
Current dataset statistics as JSON
- /mcptext/html
How agents query this dataset over MCP
- /api/mcpapplication/json
Machine-readable MCP server descriptor
- /catalog.mdtext/markdown
Every indexable URL with its markdown rendition
- /api/catalogapplication/json
Available JSON endpoints and their shapes
- /auth.mdtext/markdown
Access terms for automated clients