One file. Your agent reads your plan · and proposes changes you approve.

Install · 3 steps

1. Step 1 · mint a token

   Open Settings → Alignment for AI agents and mint a token with a memorable name (e.g. Claude Desktop). Copy the pna_… value · it's shown once. PathNav stores only a hash.

2. Step 2 · save the script

   Save pathnav-mcp.mjs somewhere stable (e.g. ~/.local/bin/pathnav-mcp.mjs). Source is small enough to read in 2 minutes; zero npm deps.

   Alternative · curl directly:

   curl -L https://pathnav.ai/mcp/pathnav-mcp.mjs -o pathnav-mcp.mjs

3. Step 3 · register with your MCP host

   Claude Desktop · ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%/Claude/claude_desktop_config.json on Windows:

   {
     "mcpServers": {
       "pathnav": {
         "command": "node",
         "args": ["/absolute/path/to/pathnav-mcp.mjs"],
         "env": {
           "PATHNAV_TOKEN": "pna_..."
         }
       }
     }
   }

   Cursor · Settings → MCP → Add server, same JSON shape.

   Restart the host. Your agent now sees PathNav tools and will call get_alignment on its own when relevant · and propose_change when it has something concrete worth recording (with your approval).

Tools the agent gains

• get_alignment · read-only

  Full JSON · pillars, vision, goals + milestones, drains (anti-goals), parked, projects, habits, current focus.

• get_alignment_markdown · read-only

  Same data as a markdown brief. Prefer this when the agent needs to quote your plan back to you.

• propose_change · destructive, requires approval

  Submit a milestone, note, drain, focus, or parked item. Lands in your review queue · you approve or reject. Never auto-applies.

• list_pending_proposals · read-only

  See what’s already in your review queue. Agents check this before submitting so duplicates don’t pile up.

• get_relevant_knowledge · read-only

  Retrieve approved newsletter, knowledge-pack, Git-backed, document, or MCP-resource guidance by task, pillar, goal, trust, freshness, and agent scope.

• get_visual_primitives · read-only

  Read the safe blank-canvas primitive catalog before storing a PathNav-rendered visual.interface.v1 design.

• list_objects / get_object · read-only

  Inspect shared memory, documents, policies, monitors, visual designs, prior actions, and reconciliation findings.

• store_object · propose scope

  Store typed records, documents, metrics, monitors, policies, or actions. Every write runs reconciliation against the user's goals and guardrails.

• health · read-only

  Sanity check · verifies the token + PathNav reachability.