spooling 0.2.11__tar.gz → 0.2.13__tar.gz

{spooling-0.2.11/spooling.egg-info → spooling-0.2.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: spooling
-Version: 0.2.11
+Version: 0.2.13
 Summary: Local session tracker and semantic search for AI coding assistants
 Author: Parsed Analytics, Inc.
 License: MIT
@@ -155,28 +155,42 @@ spooling stats --week      # Weekly breakdown
 spooling stats --days 30   # Last 30 days
 ```
 
-### `spooling tools`, `spooling agents`, `spooling traces`
+### `spooling tools`, `spooling agents`, `spooling mcps`, `spooling traces`
 
-See which tools and agents your sessions use. These read the spans stored by `spooling sync`, and use the same agent/tool names as the web app (an agent is its subagent type, falling back to its span name).
+See which tools, MCP servers and agents your sessions use. These read the spans stored by `spooling sync`, and use the same names as the web app (an agent is its subagent type, falling back to its span name; an MCP tool is named `mcp__<server>__<tool>`).
 
 ```bash
 spooling tools                       # Top tools: uses, traces, errors, avg time
+spooling tools -s read               # Tools whose name contains "read"
 spooling tools --agent explore       # Tools called by the "explore" agent
-spooling tools --days 7 --provider claude-code
+spooling tools --mcp github          # Tools of MCP servers matching "github"
+
+spooling mcps                        # Every MCP server used: tools, calls, traces, errors
+spooling mcps spool                  # The tools of servers matching "spool"
 
 spooling agents                      # Agents: runs, traces, tokens, cost
+spooling agents -s expl              # Agents whose name contains "expl"
 spooling agents --tool Bash          # Agents whose runs called Bash
+spooling agents --mcp github         # Agents whose runs used a GitHub MCP tool
 
-spooling traces                      # Latest traces, with the agents each ran
-spooling traces ses_10dd             # Search title, project, session id, trace id or agent name
-spooling traces --agent explore --tool Bash --days 30
+spooling traces                      # Latest traces, with the agents and MCP servers each used
+spooling traces ses_10dd             # Search title, project, session id, trace id, agent or MCP server name
+spooling traces --agent explore --tool Bash --mcp spool --days 30
 ```
 
-Common options: `--days N`, `--project TEXT`, `--provider/-p NAME`, `--limit/-n N`, and `--json` for scripting. `--agent` and `--tool` can be repeated (matches any).
+Common options: `--days N`, `--project TEXT`, `--provider/-p NAME`, `--limit/-n N`, and `--json` for scripting.
+
+Matching rules: `--agent` and `--tool` match the exact name, ignoring case, and can be repeated (matches any). `--mcp` and `--search/-s` match part of the name, ignoring case. When you combine different filters on `traces`, a trace must satisfy all of them.
+
+`tools --agent` and `agents --tool/--mcp` look inside an agent's own runs. `traces --agent/--tool/--mcp` matches whole traces, like the Traces page.
+
+`--cloud` on `tools`, `agents` and `mcps` shows workspace-wide totals from Spooling Cloud. The cloud endpoints only return the 100 most-used tools, so `--search` and `--mcp` filter those client-side, and `--days/--project/--provider/--agent/--tool` can't be combined with `--cloud`.
+
+Agent data is only recorded for providers that expose subagents (currently Claude Code, via its `Agent` tool, previously called `Task`), so `spooling agents` is empty for other providers. MCP calls are recognised by the `mcp__<server>__<tool>` name Claude Code uses; tools from other clients that name MCP tools differently (for example `server_tool`) appear in `spooling tools` but not in `spooling mcps`.
 
-`tools --agent` and `agents --tool` look inside an agent's own runs. `traces --agent/--tool` matches whole traces, like the Traces page.
+After upgrading, `spooling sync` re-reads your Claude Code sessions once to pick up agents. To force a re-read of any provider's files at any time, run `spooling sync --reparse` (optionally with `-p claude-code`).
 
-`spooling tools --cloud` and `spooling agents --cloud` show workspace-wide totals from Spooling Cloud. The cloud endpoints only support `--limit`, so combining `--cloud` with other filters is an error. Agent data is only recorded for providers that expose subagents (for example Claude Code's `Task` tool), so `spooling agents` can be empty for other providers.
+Agent runs show duration and the prompt that started them. Their own tokens, cost and tool calls are not attributed yet, because Claude Code writes subagent activity to separate transcript files that Spooling does not read yet.
 
 ### `spooling search <query>`
 

{spooling-0.2.11 → spooling-0.2.13}/README.md

@@ -119,28 +119,42 @@ spooling stats --week      # Weekly breakdown
 spooling stats --days 30   # Last 30 days
 ```
 
-### `spooling tools`, `spooling agents`, `spooling traces`
+### `spooling tools`, `spooling agents`, `spooling mcps`, `spooling traces`
 
-See which tools and agents your sessions use. These read the spans stored by `spooling sync`, and use the same agent/tool names as the web app (an agent is its subagent type, falling back to its span name).
+See which tools, MCP servers and agents your sessions use. These read the spans stored by `spooling sync`, and use the same names as the web app (an agent is its subagent type, falling back to its span name; an MCP tool is named `mcp__<server>__<tool>`).
 
 ```bash
 spooling tools                       # Top tools: uses, traces, errors, avg time
+spooling tools -s read               # Tools whose name contains "read"
 spooling tools --agent explore       # Tools called by the "explore" agent
-spooling tools --days 7 --provider claude-code
+spooling tools --mcp github          # Tools of MCP servers matching "github"
+
+spooling mcps                        # Every MCP server used: tools, calls, traces, errors
+spooling mcps spool                  # The tools of servers matching "spool"
 
 spooling agents                      # Agents: runs, traces, tokens, cost
+spooling agents -s expl              # Agents whose name contains "expl"
 spooling agents --tool Bash          # Agents whose runs called Bash
+spooling agents --mcp github         # Agents whose runs used a GitHub MCP tool
 
-spooling traces                      # Latest traces, with the agents each ran
-spooling traces ses_10dd             # Search title, project, session id, trace id or agent name
-spooling traces --agent explore --tool Bash --days 30
+spooling traces                      # Latest traces, with the agents and MCP servers each used
+spooling traces ses_10dd             # Search title, project, session id, trace id, agent or MCP server name
+spooling traces --agent explore --tool Bash --mcp spool --days 30
 ```
 
-Common options: `--days N`, `--project TEXT`, `--provider/-p NAME`, `--limit/-n N`, and `--json` for scripting. `--agent` and `--tool` can be repeated (matches any).
+Common options: `--days N`, `--project TEXT`, `--provider/-p NAME`, `--limit/-n N`, and `--json` for scripting.
+
+Matching rules: `--agent` and `--tool` match the exact name, ignoring case, and can be repeated (matches any). `--mcp` and `--search/-s` match part of the name, ignoring case. When you combine different filters on `traces`, a trace must satisfy all of them.
+
+`tools --agent` and `agents --tool/--mcp` look inside an agent's own runs. `traces --agent/--tool/--mcp` matches whole traces, like the Traces page.
+
+`--cloud` on `tools`, `agents` and `mcps` shows workspace-wide totals from Spooling Cloud. The cloud endpoints only return the 100 most-used tools, so `--search` and `--mcp` filter those client-side, and `--days/--project/--provider/--agent/--tool` can't be combined with `--cloud`.
+
+Agent data is only recorded for providers that expose subagents (currently Claude Code, via its `Agent` tool, previously called `Task`), so `spooling agents` is empty for other providers. MCP calls are recognised by the `mcp__<server>__<tool>` name Claude Code uses; tools from other clients that name MCP tools differently (for example `server_tool`) appear in `spooling tools` but not in `spooling mcps`.
 
-`tools --agent` and `agents --tool` look inside an agent's own runs. `traces --agent/--tool` matches whole traces, like the Traces page.
+After upgrading, `spooling sync` re-reads your Claude Code sessions once to pick up agents. To force a re-read of any provider's files at any time, run `spooling sync --reparse` (optionally with `-p claude-code`).
 
-`spooling tools --cloud` and `spooling agents --cloud` show workspace-wide totals from Spooling Cloud. The cloud endpoints only support `--limit`, so combining `--cloud` with other filters is an error. Agent data is only recorded for providers that expose subagents (for example Claude Code's `Task` tool), so `spooling agents` can be empty for other providers.
+Agent runs show duration and the prompt that started them. Their own tokens, cost and tool calls are not attributed yet, because Claude Code writes subagent activity to separate transcript files that Spooling does not read yet.
 
 ### `spooling search <query>`
 

{spooling-0.2.11 → spooling-0.2.13}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "spooling"
-version = "0.2.11"
+version = "0.2.13"
 description = "Local session tracker and semantic search for AI coding assistants"
 readme = "README.md"
 authors = [{ name = "Parsed Analytics, Inc." }]

{spooling-0.2.11 → spooling-0.2.13}/spooling/cli.py

@@ -65,10 +65,11 @@ def init():
 @cli.command()
 @click.option("--no-embed", is_flag=True, help="Skip embedding (faster sync)")
 @click.option("--provider", "-p", default=None, help="Only sync a specific provider (claude-code, jsonl-session, codex, cursor, copilot, windsurf)")
-def sync(no_embed, provider):
+@click.option("--reparse", is_flag=True, help="Re-read session files even if they look unchanged (rebuilds traces/spans)")
+def sync(no_embed, provider, reparse):
     """Sync AI coding sessions to the database."""
     from spooling.ingest import sync as do_sync
-    do_sync(embed=not no_embed, provider_filter=provider)
+    do_sync(embed=not no_embed, provider_filter=provider, reparse=reparse)
 
 
 @cli.command()
@@ -1372,7 +1373,8 @@ def _reject_cloud_filters(**filters) -> None:
     if used:
         raise click.UsageError(
             f"{', '.join(used)} can't be combined with --cloud "
-            "(the cloud endpoint only supports --limit). Drop --cloud to filter local data."
+            "(the cloud endpoint only supports --limit, --search and --mcp). "
+            "Drop --cloud to filter local data."
         )
 
 
@@ -1391,23 +1393,44 @@ def _filter_opts(fn):
     return fn
 
 
+def _error_cell(n) -> str:
+    return f"[red]{n}[/red]" if n else "0"
+
+
+def _cloud_tool_rows(limit: int, search: str | None, mcps) -> list[dict]:
+    """Cloud per-tool rows, filtered client-side by name/MCP server (server caps at 100)."""
+    from spooling.cloud import cloud_tools
+    from spooling.insights import mcp_server_of
+    rows = cloud_tools(100)
+    if search:
+        rows = [r for r in rows if search.lower() in r["tool_name"].lower()]
+    if mcps:
+        rows = [r for r in rows
+                if any(m.lower() in (mcp_server_of(r["tool_name"]) or "").lower() for m in mcps)]
+    return rows[:limit]
+
+
 @cli.command("tools")
 @click.option("--agent", "agents", multiple=True, help="Only tools called by this agent (repeatable)")
+@click.option("--mcp", "mcps", multiple=True, help="Only tools of MCP servers whose name contains this (repeatable)")
+@click.option("--search", "-s", default=None, help="Only tools whose name contains this text")
 @click.option("--cloud", "cloud_mode", is_flag=True, help="Use the cloud workspace instead of local data")
 @_filter_opts
-def tools_cmd(agents, cloud_mode, days, project, provider, limit, as_json):
+def tools_cmd(agents, mcps, search, cloud_mode, days, project, provider, limit, as_json):
     """Show which tools are used, how often, and how often they fail.
 
+    MCP tools are named mcp__<server>__<tool>; use --mcp to see one server's tools.
+
     \b
     spooling tools                      top tools across all traces
+    spooling tools -s read              tools whose name contains "read"
+    spooling tools --mcp spool          tools from MCP servers matching "spool"
     spooling tools --agent explore      tools called by the 'explore' agent
-    spooling tools --days 7 --json
-    spooling tools --cloud              workspace-wide, from Spooling Cloud
+    spooling tools --cloud --mcp github workspace-wide, from Spooling Cloud
     """
     if cloud_mode:
         _reject_cloud_filters(agent=agents, days=days, project=project, provider=provider)
-        from spooling.cloud import cloud_tools
-        rows = cloud_tools(limit)
+        rows = _cloud_tool_rows(limit, search, mcps)
         if as_json:
             return _emit_json(rows)
         table = Table(title="Tools (cloud)", show_lines=False)
@@ -1416,20 +1439,25 @@ def tools_cmd(agents, cloud_mode, days, project, provider, limit, as_json):
         table.add_column("Sessions", justify="right")
         table.add_column("Errors", justify="right")
         for r in rows:
-            table.add_row(r["tool_name"], _fmt_num(r["uses"]), _fmt_num(r["sessions"]),
-                          f"[red]{r['errors']}[/red]" if r["errors"] else "0")
-        console.print(table if rows else "[yellow]No tool data in the cloud workspace yet.[/yellow]")
+            table.add_row(r["tool_name"], _fmt_num(r["uses"]), _fmt_num(r["sessions"]), _error_cell(r["errors"]))
+        console.print(table if rows else "[yellow]No matching tools in the cloud workspace.[/yellow]")
         return
 
     from spooling.insights import tool_stats
-    rows = tool_stats(days=days, project=project, provider=provider, agents=list(agents) or None, limit=limit)
+    rows = tool_stats(days=days, project=project, provider=provider, agents=list(agents) or None,
+                      mcps=list(mcps) or None, search=search, limit=limit)
     if as_json:
         return _emit_json(rows)
     if not rows:
-        console.print("[yellow]No tool data found. Run 'spooling sync' first"
+        console.print("[yellow]No matching tool data found. Run 'spooling sync' first"
                       + (" (agent filters need traces that contain agent spans)." if agents else ".") + "[/yellow]")
         return
-    table = Table(title="Tools" + (f" called by {', '.join(agents)}" if agents else ""), show_lines=False)
+    title = "Tools"
+    if agents:
+        title += f" called by {', '.join(agents)}"
+    if mcps:
+        title += f" from MCP {', '.join(mcps)}"
+    table = Table(title=title, show_lines=False)
     table.add_column("Tool", style="cyan")
     table.add_column("Uses", justify="right")
     table.add_column("Traces", justify="right")
@@ -1437,27 +1465,130 @@ def tools_cmd(agents, cloud_mode, days, project, provider, limit, as_json):
     table.add_column("Avg time", justify="right", style="dim")
     for r in rows:
         table.add_row(r["name"], _fmt_num(r["uses"]), _fmt_num(r["traces"]),
-                      f"[red]{r['errors']}[/red]" if r["errors"] else "0", _fmt_ms(r["avg_ms"]))
+                      _error_cell(r["errors"]), _fmt_ms(r["avg_ms"]))
     console.print(table)
 
 
+@cli.command("mcps")
+@click.argument("server", required=False)
+@click.option("--search", "-s", default=None, help="Only servers/tools whose name contains this text")
+@click.option("--cloud", "cloud_mode", is_flag=True, help="Use the cloud workspace instead of local data")
+@_filter_opts
+def mcps_cmd(server, search, cloud_mode, days, project, provider, limit, as_json):
+    """Show which MCP servers were used, and which of their tools.
+
+    With no argument, one row per MCP server. Pass a server name (or part of
+    it) to list that server's tools instead.
+
+    \b
+    spooling mcps                       every MCP server used: tools, calls, traces, errors
+    spooling mcps spool                 the tools of servers matching "spool"
+    spooling mcps -s search             servers/tools with "search" in the name
+    spooling mcps --days 7 --json
+    spooling mcps --cloud               workspace-wide, from Spooling Cloud
+    """
+    if cloud_mode:
+        _reject_cloud_filters(days=days, project=project, provider=provider)
+        from spooling.insights import group_mcp_tools
+        rows = _cloud_tool_rows(100, search, [server] if server else None)
+        mcp_rows = [r for r in rows if r["tool_name"].lower().startswith("mcp__")]
+        if server:
+            mcp_rows = mcp_rows[:limit]
+            if as_json:
+                return _emit_json(mcp_rows)
+            table = Table(title=f"MCP tools matching '{server}' (cloud)", show_lines=False)
+            table.add_column("Tool", style="cyan")
+            table.add_column("Uses", justify="right")
+            table.add_column("Sessions", justify="right")
+            table.add_column("Errors", justify="right")
+            for r in mcp_rows:
+                table.add_row(r["tool_name"], _fmt_num(r["uses"]), _fmt_num(r["sessions"]), _error_cell(r["errors"]))
+            console.print(table if mcp_rows else "[yellow]No matching MCP tools in the cloud workspace.[/yellow]")
+            return
+        grouped = group_mcp_tools(mcp_rows)[:limit]
+        if as_json:
+            return _emit_json(grouped)
+        table = Table(title="MCP servers (cloud)", show_lines=False)
+        table.add_column("MCP server", style="cyan")
+        table.add_column("Tools", justify="right")
+        table.add_column("Calls", justify="right")
+        table.add_column("Errors", justify="right")
+        for g in grouped:
+            table.add_row(g["name"], _fmt_num(g["tools"]), _fmt_num(g["calls"]), _error_cell(g["errors"]))
+        console.print(table if grouped else "[yellow]No MCP tool calls in the cloud workspace.[/yellow]")
+        console.print("[dim]Based on the 100 most-used tools in the workspace.[/dim]")
+        return
+
+    from spooling.insights import mcp_stats, tool_stats
+    if server:
+        rows = tool_stats(days=days, project=project, provider=provider, mcps=[server],
+                          search=search, limit=limit)
+        if as_json:
+            return _emit_json(rows)
+        if not rows:
+            console.print(f"[yellow]No MCP tools found for a server matching '{server}'. "
+                          "Run 'spooling mcps' to see the servers.[/yellow]")
+            return
+        table = Table(title=f"MCP tools matching '{server}'", show_lines=False)
+        table.add_column("Tool", style="cyan")
+        table.add_column("Server", style="dim")
+        table.add_column("Uses", justify="right")
+        table.add_column("Traces", justify="right")
+        table.add_column("Errors", justify="right")
+        table.add_column("Avg time", justify="right", style="dim")
+        for r in rows:
+            table.add_row(r["name"], r["mcp_server"] or "", _fmt_num(r["uses"]), _fmt_num(r["traces"]),
+                          _error_cell(r["errors"]), _fmt_ms(r["avg_ms"]))
+        console.print(table)
+        return
+
+    rows = mcp_stats(days=days, project=project, provider=provider, search=search, limit=limit)
+    if as_json:
+        return _emit_json(rows)
+    if not rows:
+        console.print(
+            "[yellow]No MCP tool calls found.[/yellow]\n"
+            "[dim]MCP tools are recorded as mcp__<server>__<tool>. Run 'spooling sync' after using one, "
+            "or try 'spooling mcps --cloud'.[/dim]"
+        )
+        return
+    table = Table(title="MCP servers", show_lines=False)
+    table.add_column("MCP server", style="cyan")
+    table.add_column("Tools", justify="right")
+    table.add_column("Calls", justify="right")
+    table.add_column("Traces", justify="right")
+    table.add_column("Errors", justify="right")
+    table.add_column("Avg time", justify="right", style="dim")
+    for r in rows:
+        table.add_row(r["name"], _fmt_num(r["tools"]), _fmt_num(r["calls"]), _fmt_num(r["traces"]),
+                      _error_cell(r["errors"]), _fmt_ms(r["avg_ms"]))
+    console.print(table)
+    console.print("[dim]spooling mcps <server> lists a server's tools.[/dim]")
+
+
 @cli.command("agents")
 @click.option("--tool", "tools", multiple=True, help="Only agents that called this tool (repeatable)")
+@click.option("--mcp", "mcps", multiple=True, help="Only agents that called a tool of an MCP server matching this (repeatable)")
+@click.option("--search", "-s", default=None, help="Only agents whose name contains this text")
 @click.option("--cloud", "cloud_mode", is_flag=True, help="Use the cloud workspace instead of local data")
 @_filter_opts
-def agents_cmd(tools, cloud_mode, days, project, provider, limit, as_json):
+def agents_cmd(tools, mcps, search, cloud_mode, days, project, provider, limit, as_json):
     """Show which agents (subagent types) ran, how often, and what they cost.
 
     \b
     spooling agents                     agents across all traces
+    spooling agents -s expl             agents whose name contains "expl"
     spooling agents --tool Bash         agents whose runs called Bash
-    spooling agents --days 30 --json
+    spooling agents --mcp github        agents whose runs called a GitHub MCP tool
     spooling agents --cloud             workspace-wide, from Spooling Cloud
     """
     if cloud_mode:
-        _reject_cloud_filters(tool=tools, days=days, project=project, provider=provider)
+        _reject_cloud_filters(tool=tools, mcp=mcps, days=days, project=project, provider=provider)
         from spooling.cloud import cloud_agents
-        rows = cloud_agents()[:limit]
+        rows = cloud_agents()
+        if search:
+            rows = [r for r in rows if search.lower() in r["agent_type"].lower()]
+        rows = rows[:limit]
         if as_json:
             return _emit_json(rows)
         table = Table(title="Agents (cloud)", show_lines=False)
@@ -1465,21 +1596,28 @@ def agents_cmd(tools, cloud_mode, days, project, provider, limit, as_json):
         table.add_column("Runs", justify="right")
         for r in rows:
             table.add_row(r["agent_type"], _fmt_num(r["uses"]))
-        console.print(table if rows else "[yellow]No agent data in the cloud workspace yet.[/yellow]")
+        console.print(table if rows else "[yellow]No matching agents in the cloud workspace.[/yellow]")
         return
 
     from spooling.insights import agent_stats
-    rows = agent_stats(days=days, project=project, provider=provider, tools=list(tools) or None, limit=limit)
+    rows = agent_stats(days=days, project=project, provider=provider, tools=list(tools) or None,
+                       mcps=list(mcps) or None, search=search, limit=limit)
     if as_json:
         return _emit_json(rows)
     if not rows:
         console.print(
-            "[yellow]No agent spans found locally.[/yellow]\n"
-            "[dim]Agent runs are recorded for providers that expose subagents (e.g. Claude Code's Task tool). "
-            "Run 'spooling sync' after using one, or try 'spooling agents --cloud'.[/dim]"
+            "[yellow]No matching agent spans found locally.[/yellow]\n"
+            "[dim]Agent runs are recorded for providers that expose subagents (currently Claude Code, via its "
+            "Agent/Task tool). Run 'spooling sync' after using one, or try 'spooling agents --cloud'.\n"
+            "If you use Claude Code and still see this, run 'spooling sync --reparse -p claude-code'.[/dim]"
         )
         return
-    table = Table(title="Agents" + (f" that called {', '.join(tools)}" if tools else ""), show_lines=False)
+    title = "Agents"
+    if tools:
+        title += f" that called {', '.join(tools)}"
+    if mcps:
+        title += f" that used MCP {', '.join(mcps)}"
+    table = Table(title=title, show_lines=False)
     table.add_column("Agent", style="cyan")
     table.add_column("Runs", justify="right")
     table.add_column("Traces", justify="right")
@@ -1496,22 +1634,26 @@ def agents_cmd(tools, cloud_mode, days, project, provider, limit, as_json):
 @click.argument("query", required=False)
 @click.option("--agent", "agents", multiple=True, help="Only traces that ran this agent (repeatable)")
 @click.option("--tool", "tools", multiple=True, help="Only traces that used this tool (repeatable)")
+@click.option("--mcp", "mcps", multiple=True, help="Only traces that used an MCP server matching this (repeatable)")
 @click.option("--session", default=None, help="Session id (or unique prefix)")
 @_filter_opts
-def traces_cmd(query, agents, tools, session, days, project, provider, limit, as_json):
-    """List traces, optionally filtered by agent, tool, or free text.
+def traces_cmd(query, agents, tools, mcps, session, days, project, provider, limit, as_json):
+    """List traces, optionally filtered by agent, tool, MCP server, or free text.
 
-    QUERY matches title, project, session id, trace id or an agent name.
+    QUERY matches title, project, session id, trace id, an agent name or an MCP
+    server name. --agent and --tool match the exact name, ignoring case.
 
     \b
     spooling traces                           latest traces
     spooling traces ses_10dd                  find by session id
     spooling traces --agent explore           traces that ran the explore agent
+    spooling traces --mcp spool               traces that used an MCP server matching "spool"
     spooling traces --tool Bash --days 7      traces that used Bash this week
     """
     from spooling.insights import list_traces
     rows = list_traces(query=query, agents=list(agents) or None, tools=list(tools) or None,
-                       session=session, days=days, project=project, provider=provider, limit=limit)
+                       mcps=list(mcps) or None, session=session, days=days, project=project,
+                       provider=provider, limit=limit)
     if as_json:
         return _emit_json(rows)
     if not rows:
@@ -1522,17 +1664,21 @@ def traces_cmd(query, agents, tools, session, days, project, provider, limit, as
     table.add_column("Title", style="cyan", max_width=22, overflow="ellipsis", no_wrap=True)
     table.add_column("Session", style="dim", no_wrap=True)
     table.add_column("Agents", max_width=12, overflow="ellipsis", no_wrap=True)
+    table.add_column("MCP", max_width=12, overflow="ellipsis", no_wrap=True)
     table.add_column("Spans", justify="right", no_wrap=True)
     table.add_column("Tools", justify="right", no_wrap=True)
     table.add_column("Cost", justify="right", no_wrap=True)
+
+    def short(names: list[str]) -> str:
+        return ", ".join(names[:1]) + (f" +{len(names) - 1}" if len(names) > 1 else "") if names else "-"
+
     for r in rows:
-        ag = r["agents"]
-        ag_txt = ", ".join(ag[:1]) + (f" +{len(ag) - 1}" if len(ag) > 1 else "") if ag else "-"
         table.add_row(
             str(r["started_at"] or "")[5:10],
             r["title"] or "(untitled)",
             (r["session_id"] or "")[:10],
-            ag_txt,
+            short(r["agents"]),
+            short(r["mcps"]),
             _fmt_num(r["span_count"]),
             _fmt_num(r["tool_count"]),
             f"${float(r['total_cost_usd'] or 0):.2f}",

{spooling-0.2.11 → spooling-0.2.13}/spooling/ingest.py

@@ -458,8 +458,42 @@ def _sync_remote_provider(conn, provider, prov_info: dict, embed: bool) -> tuple
     return total_sessions, total_messages, total_chunks
 
 
-def sync(embed: bool = True, provider_filter: str | None = None):
-    """Sync sessions from all connected providers to the database."""
+_REVISION_PREFIX = "__parser_revision__:"
+
+
+def _needs_reread(conn, synced: dict, provider_type: str, revision: int, reparse: bool) -> bool:
+    """True if this provider's already-synced files must be re-read this run.
+
+    Either the user asked (``--reparse``), or a parser fix shipped
+    (``provider.parser_revision`` is newer than the one recorded at last sync)
+    and there are previously synced files that were parsed the old way.
+    """
+    if reparse:
+        return True
+    if not revision:
+        return False
+    stored = synced.get(_REVISION_PREFIX + provider_type)
+    if stored is not None and stored >= revision:
+        return False
+    row = conn.execute(
+        "SELECT COUNT(*) AS n FROM sync_state WHERE provider_id = %s AND file_path NOT LIKE %s",
+        (provider_type, _REVISION_PREFIX + "%"),
+    ).fetchone()
+    return bool(row["n"])
+
+
+def _record_revision(conn, synced: dict, provider_type: str, revision: int) -> None:
+    if revision and synced.get(_REVISION_PREFIX + provider_type) != revision:
+        _mark_synced(conn, _REVISION_PREFIX + provider_type, revision, provider_type)
+        synced[_REVISION_PREFIX + provider_type] = revision
+
+
+def sync(embed: bool = True, provider_filter: str | None = None, reparse: bool = False):
+    """Sync sessions from all connected providers to the database.
+
+    ``reparse`` re-reads files even if they look unchanged (also done
+    automatically, once, when a provider's parser improves).
+    """
     conn = get_connection()
     connected = _get_connected_providers(conn)
 
@@ -533,20 +567,31 @@ def sync(embed: bool = True, provider_filter: str | None = None):
         if not files:
             continue
 
+        revision = getattr(provider, "parser_revision", 0)
+        reread = _needs_reread(conn, synced, prov_info["type"], revision, reparse)
+        # When re-reading, pretend nothing was synced so every file is parsed again.
+        prov_synced = {} if reread else synced
+
         # Filter to new or changed files. Incremental providers (many sessions
         # per file) are always scanned; they skip unchanged sessions themselves.
         to_process = []
         for f in files:
             size = f.stat().st_size
-            if provider.incremental or str(f) not in synced or synced[str(f)] != size:
+            if provider.incremental or str(f) not in prov_synced or prov_synced[str(f)] != size:
                 to_process.append(f)
 
         if not to_process:
+            _record_revision(conn, synced, prov_info["type"], revision)
+            conn.commit()
             continue
 
         console.print(
             f"[bold]{provider.name}:[/bold] Found {len(to_process)} new/updated session files."
         )
+        if reread and not reparse:
+            console.print(
+                f"  [dim]Re-reading {provider.name} sessions once to pick up improved parsing.[/dim]"
+            )
 
         total_messages = 0
         total_chunks = 0
@@ -570,7 +615,7 @@ def sync(embed: bool = True, provider_filter: str | None = None):
                 batch_sessions: list[tuple] = []  # (file, [sessions])
 
                 for f in batch_files:
-                    sessions = parse_file(provider, f, synced)
+                    sessions = parse_file(provider, f, prov_synced)
                     for session in sessions:
                         _store_session(conn, session)
                         if session.trace is not None:
@@ -601,6 +646,8 @@ def sync(embed: bool = True, provider_filter: str | None = None):
                     _mark_synced(conn, str(f), f.stat().st_size, prov_info["type"])
                 conn.commit()
 
+        _record_revision(conn, synced, prov_info["type"], revision)
+
         # Update provider stats
         conn.execute(
             """UPDATE providers SET