Ecosystem Intelligence MCP

A sellable prototype for bounded, read-only understanding of local project directories. It is designed to help an AI assistant inventory projects, identify technology markers, and read selected text files without executing code or exposing arbitrary filesystem paths.

Product documents

• What This Is

• What This Can Do

• Changelog

• Advanced Handoff

• Build and Review Record

• Next Phase Specification

• Session Checkpoint

Related MCP server: MCP Workspace Server

Tools

• inventory_roots — shows the configured allowlist.

• summarize_project — bounded directory structure, file counts, extensions, markers, and representative text-file paths.

• read_project_file — reads a small recognized text/project file under an approved root with sensitive-path filtering and redaction.

• analyze_project — infers purpose and capabilities from bounded evidence, with confidence, readiness, unknowns, and safety metadata.

It also exposes ecosystem://policy as a resource describing the safety model.

Run locally

npm install
ECOSYSTEM_ROOTS="$HOME/Documents" npm start

The server communicates over MCP stdio. It does not execute project code, mutate files, follow symlink escapes, inspect unconfigured roots, or read arbitrary binary files. Common sensitive paths are excluded and common secret patterns are redacted.

Package direction

manifest.json is MCPB-ready for a future downloadable bundle. Before selling, add signed release artifacts, a privacy policy, support documentation, a second platform test, and a hosted/paid tier only after validating demand.

Product boundary

The commercial moat should be the interpretation layer: project-purpose extraction, capability overlap, maturity scoring, and actionable productization recommendations. Keep those higher-level features separate from the local filesystem adapter so the product can later support uploaded manifests, Git repositories, or a hosted service.