History

Natalie 2d5ae395dc feat(mcp): add classify_message tool to mcp-prospector (prospector copilot) + architecture docs for desktop coworker composition mcp-prospector is the prospector copilot MCP for Claude desktop coworker (Executor) to use central intelligence. - New tool `classify_message(text, handle?, phone?, has_call?, local_is_known_contact?)`: runs fast efficient rules classifier (our distilled model from training data) + full central stack (LLM via GPU model-boss if deployed, mrnumber, macsync/quinn-messenger context via thread/classify, qualification). Supports passing local desktop context for Contacts gate. - Local fastClassify/fastTemplate inlined for the MCP (pure, no extra dep issues). - Updated README with full architecture: - Composition: quinn-messenger (messaging), mrnumber (screening), classifier (fast + GPU LLM), macsync (data), quinn-desktop (local MCP for addressbook/Contacts gate, other sensors). - Desktop coworker loads multiple MCPs (this central + quinn-messenger MCP + quinn-desktop local MCP). - SKILL instructs workflow: desktop local lookup first, then central classify_message for combined result using optimized model + all sources. - Replaces pure local stopgap rules with central quality while keeping desktop facts. - This lets the desktop coworker classify messages using the full prospector copilot feature (central brain on GPU + local). Follows project: prospector central in api, macsync network, MCP as tool interface for agents, hybrid fast rules + LLM. Part of making prospector copilot the way for coworker + replace claude deps (GPU for LLM part).		2026-06-28 11:59:10 -04:00
..
src	feat(mcp): add classify_message tool to mcp-prospector (prospector copilot) + architecture docs for desktop coworker composition	2026-06-28 11:59:10 -04:00
bun.lock
package.json	refactor(mcp-prospector): delegate to @lilith/agent-prospector	2026-06-21 17:35:08 -05:00
README.md	feat(mcp): add classify_message tool to mcp-prospector (prospector copilot) + architecture docs for desktop coworker composition	2026-06-28 11:59:10 -04:00
tsconfig.json

README.md

quinn-prospector MCP

Black quinn.api corpus + /my/prospects cockpit API (same as quinn.my Prospector).

Setup

quinn-prospector → http://black.lan:3912/mcp (.mcp.json)

Redeploy: deployments/@domains/quinn.mcp/deploy.sh prospector

Cockpit workflow

cockpit_stream
cockpit_tour_cohort(city="New York (Manhattan)")
cockpit_announce(city="New York (Manhattan)")
cockpit_thread(handle="+1…")
cockpit_draft(handle="+1…")
cockpit_send(handle="+1…", message="…", draft_id=…)

Tools

Cockpit (UI parity): cockpit_stream, cockpit_tour_board, cockpit_tour_cohort, cockpit_announce, cockpit_announce_status, cockpit_backfill_status, cockpit_backfill_start, cockpit_classify, cockpit_thread, cockpit_draft, cockpit_list_drafts, cockpit_send, cockpit_correction, cockpit_correction_patterns, cockpit_corrections_recent, cockpit_mark_worked

Corpus: get_prospect, list_prospects, list_by_archetype, get_metro_demand, tag_prospect

New for desktop coworker / copilot composition: classify_message (see below), plus existing mr_number_* for screening.

Architecture for prospector copilot (Claude desktop coworker + quinn-desktop + quinn-messenger + macsync + classifier + mrnumber)

The mcp-prospector is the prospector copilot MCP — the composition point for prospect intelligence.

quinn-messenger / macsync: threads, sends, calls, notes (Pastebin) via the cockpit/thread tools and backend.
mrnumber: screening via mr_number_check / history tools (record app lookups; auto reputation).
classifier: full via cockpit_classify (central, uses fast rules + LLM on GPU via model-boss when deployed) + classify_message for raw/single message.
quinn-desktop: separate local MCP (desktop-side osascript/Contacts for hard addressbook gate, local files, TTS, etc.). Desktop coworker loads both this central MCP and the quinn-desktop MCP. Agent reasoning composes: call desktop MCP for local_is_known_contact / addressbook on phone, then pass to central classify_message or use in SKILL before central classify.
Classifier fast path: the efficient rules model (distilled from training set) runs locally in this MCP for instant classification; central LLM (GPU) for full/ambiguous via the prospect backend.

For Claude desktop coworker (Executor sessions):

Load mcp-prospector (central prospector copilot) + quinn-messenger MCP + quinn-desktop MCP (local).
In the SKILL/prompt for inbound: "To classify a message: 1. If phone, call quinn-desktop MCP for local addressbook/contacts lookup (for the hard Contacts gate). 2. Optionally call quinn-messenger MCP for extra context. 3. Call mcp-prospector classify_message (or cockpit_classify/cockpit_thread for handle) with text + local_is_known_contact + phone. This gives combined fast + full (GPU model, mrnumber, macsync, qualification). Prefer central for quality over pure local rules."
This replaces or augments the pure local stopgap rules, giving the desktop coworker the optimized central model + all data sources while keeping local desktop facts.

classify_message(text, handle?, phone?, has_call?, local_is_known_contact?):

Runs fast efficient rules (from past-week training data + PROSPECTOR_TRAINING).
If handle: full central classify (LLM/GPU + full stack).
Includes mrnumber signals if you called mr tools or provide phone.
Pass local_is_known_contact from desktop MCP for the identity gate.
Returns fast_cat + template + full result + guidance.

This is the proper loose-coupled composition via MCP tools + agent orchestration. Central brain (prospector in api + mcp-prospector), local actuators/sensors via desktop MCP. No tight coupling, follows macsync-as-source, prospector as feature.

Redeploy after changes: deployments/@domains/quinn.mcp/deploy.sh prospector (or equivalent).