AgentCorp-Inc Exploring MCP sessions
An MCP session is one agent run, identified by $session_id on the
$mcp_tool_call event. A session is just the set of $mcp_tool_call events that
share a $session_id, ordered by timestamp. Listing sessions and reading
a session's tool calls are both plain HogQL over events; the full property
schema and recipes are in the shared reference:
products/posthog_ai/skills/querying-posthog-data/references/models-mcp.md.
The one thing SQL cannot do is summarise the agent's goal in prose — that is
the typed tool posthog:mcp-analytics-sessions-generate-intent.
Tools
| Tool | Purpose |
|---|---|
posthog:execute-sql |
List sessions and read a session's tool-call sequence (HogQL) |
posthog:mcp-analytics-sessions-generate-intent |
Generate (or fetch cached) LLM summary of a session's goal |
Workflow: list recent sessions
Group $mcp_tool_call by $session_id, deriving start/end, duration, call
count, error count, and the harness:
posthog:execute-sql
SELECT
$session_id AS session_id,
min(timestamp) AS session_start,
max(timestamp) AS session_end,
dateDiff('second', min(timestamp), max(timestamp)) AS duration_seconds,
count() AS tool_calls,
countIf(toBool(properties.$mcp_is_error)) AS errors,
any(properties.$mcp_client_name) AS client
FROM events
WHERE event = '$mcp_tool_call'
AND $session_id != ''
AND timestamp >= now() - INTERVAL 7 DAY
GROUP BY session_id
ORDER BY session_start DESC
LIMIT 50
Add HAVING errors > 0 to surface only sessions that hit failures.
Workflow: read one session's tool calls
The chronological sequence of what the agent did — the heart of debugging a run:
posthog:execute-sql
SELECT
timestamp,
coalesce(nullIf(toString(properties.$mcp_exec_tool_call_name), ''), toString(properties.$mcp_tool_name)) AS tool,
toBool(properties.$mcp_is_error) AS is_error,
toString(properties.$mcp_error_message) AS error_message,
round(toFloat(properties.$mcp_duration_ms)) AS duration_ms,
toString(properties.$mcp_intent) AS intent
FROM events
WHERE event = '$mcp_tool_call'
AND $session_id = '<session_id>'
ORDER BY timestamp ASC
Always use the effective-tool-name coalesce(...) so single-exec wrapper calls
resolve to the real tool. Read these top to bottom to reconstruct the run.
Workflow: summarise the agent's goal
When the user wants the intent of a session in prose (not the raw tool list),
call the typed tool — the first call summarises the recorded $mcp_intent
values via an LLM and persists the result; later calls return the cached summary:
posthog:mcp-analytics-sessions-generate-intent
{ "id": "<session_id>" }
Returns { "session_id": ..., "intent": "<prose summary>" }. If it returns 503,
LLM summarisation is unavailable — fall back to reading the raw $mcp_intent
values from the tool-call query above.
Constructing UI links
- Sessions list:
https://app.posthog.com/project/<project_id>/mcp-analytics/sessions
Tips
- A session with many calls but no errors that ends abruptly often means the agent gave up — check whether the last call returned a large/empty result
$mcp_intentis only present when the client supplied it; absence is common, so the generate-intent tool is the more reliable goal signal- To go from a failing tool (see
exploring-mcp-tool-quality) to the sessions that hit it, filter the sessions query on the effective tool name
Related skills
exploring-mcp-tool-quality— error rates and latency across all toolsexploring-mcp-intent-clusters— group goals across many sessions