fix(mcp): strip heavy fields from progress, remove ToolSearch, fix rename (#4953)

## Summary

- **Strip heavy fields from progress messages** — `_source_bank`,
`research`, `provenance_and_notes` were being dumped as raw JSON in
`everyrow_progress` partial rows, causing 335KB+ tool results that blew
up agent context. Now resolved citations first (needs `_source_bank`),
then stripped before serializing.
- **Remove ToolSearch from agent tools** — all tool schemas (built-in +
MCP) are loaded upfront, so deferred discovery adds no value and wastes
tokens.
- **Contextual follow-up suggestions** — system prompt now tells the
agent to suggest specific next operations based on the task output
columns, instead of generic chaining examples.
- **Fix Docker rename artifacts** — `.dockerignore` and `Dockerfile`
still referenced `everyrow-mcp/` after the `futuresearch-python` rename,
breaking local `docker compose` builds.

## Test plan

- [x] MCP server tests pass (415 passed)
- [ ] Run a rank/classify task locally and verify progress messages are
small
- [ ] Verify agent suggests contextual follow-ups after task completion
- [ ] Verify `docker compose up --build` works with the fixed Dockerfile

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Sourced from commit 6e2c84fa6761ee50eb254245cb9a799e96a3d6d0

Author: RafaelPo

.dockerignore

@@ -11,7 +11,7 @@
 !src/**
 
 # MCP server package
-!everyrow-mcp/pyproject.toml
-!everyrow-mcp/README.md
-!everyrow-mcp/src/
-!everyrow-mcp/src/**
+!futuresearch-mcp/pyproject.toml
+!futuresearch-mcp/README.md
+!futuresearch-mcp/src/
+!futuresearch-mcp/src/**

futuresearch-mcp/deploy/Dockerfile

@@ -7,11 +7,11 @@ WORKDIR /app
 # Copy workspace root pyproject + lock + README, then each package
 COPY --link pyproject.toml uv.lock README.md ./
 COPY --link src/ ./src/
-COPY --link everyrow-mcp/pyproject.toml everyrow-mcp/README.md ./everyrow-mcp/
-COPY --link everyrow-mcp/src/ ./everyrow-mcp/src/
+COPY --link futuresearch-mcp/pyproject.toml futuresearch-mcp/README.md ./futuresearch-mcp/
+COPY --link futuresearch-mcp/src/ ./futuresearch-mcp/src/
 
-# Install everyrow-mcp and its dependencies (includes workspace everyrow package)
-RUN uv sync --package everyrow-mcp --no-dev --no-sources --no-editable
+# Install futuresearch-mcp and its dependencies (includes workspace everyrow package)
+RUN uv sync --package futuresearch-mcp --no-dev --no-sources --no-editable
 
 # Stage 2: Slim runtime
 FROM python:3.13-slim
@@ -33,4 +33,4 @@ USER mcp
 HEALTHCHECK --interval=30s --timeout=5s --retries=3 \
   CMD python -c "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"
 
-CMD ["everyrow-mcp", "--http", "--port", "8000", "--host", "0.0.0.0"]
+CMD ["futuresearch-mcp", "--http", "--port", "8000", "--host", "0.0.0.0"]

futuresearch-mcp/src/futuresearch_mcp/tool_helpers.py

@@ -487,10 +487,14 @@ def progress_message(
             msg += "\n\nAgent activity:" + _format_summary_lines(summaries)
 
         if partial_rows:
+            # Resolve citations first (needs _source_bank), then strip
+            # heavy internal fields to avoid blowing up context.
+            _STRIP = {"_source_bank", "research", "provenance_and_notes"}
             resolved = resolve_citations_in_records(partial_rows)
             msg += "\n\nNewly completed rows:"
             for row in resolved:
-                msg += f"\n- {json.dumps(row, default=str)}"
+                light = {k: v for k, v in row.items() if k not in _STRIP}
+                msg += f"\n- {json.dumps(light, default=str)}"
 
         progress_call = f"futuresearch_progress(task_id='{task_id}'{cursor_arg})"