[renovate]

This PR contains the following updates:

Package	Change	Age	Confidence
@posthog/mcp (source)	^0.13.0 → ^0.25.0

---

Release Notes

PostHog/posthog-js (@​posthog/mcp)

v0.25.0

Compare Source

Minor Changes

• #​5404 68336df Thanks @​gesh! - Add a sendToolsListEvent option. It is on by default. Set it to false to stop sending $mcp_tools_list.
  (2026-10-28)

v0.24.0

Compare Source

Minor Changes

• #​5401 8faad6a Thanks @​gesh! - Add a sendInitializeEvent option. It is on by default. Set it to false to stop sending $mcp_initialize.
  (2026-10-28)

v0.23.0

Compare Source

Minor Changes

• #​5306 19b9b54 Thanks @​gesh! - Stop sending standalone $identify events, so insights, funnels, or cohorts built on MCP $identify stop receiving new data. The identify callback still sets distinct_id, $set and $groups on every event.
  (2026-10-19)

v0.22.3

Compare Source

Patch Changes

• #​5224 ec9080a Thanks @​gesh! - Set $mcp_interface to mcp on events from MCP servers.
  (2026-10-09)

v0.22.2

Compare Source

Patch Changes

• #​5162 3132998 Thanks @​pauldambra! - Sanitize each tool response once instead of twice, and replace image, audio and binary blocks before the scan, so tool calls with large results cost less CPU. Redact PostHog tokens that URL field decoding exposes.
  (2026-10-07)
• Updated dependencies [3132998]:
  • @​posthog/core@1.57.2

v0.22.1

Compare Source

Patch Changes

• #​5191 6cd5496 Thanks @​gesh! - Capture WebMCP tool intent and model metadata by default.
  (2026-10-06)
• Updated dependencies [6cd5496]:
  • @​posthog/core@1.57.0

v0.22.0

Compare Source

Minor Changes

• #​5165 3394548 Thanks @​OrbitingBucket! - Strip SDK-owned analytics arguments on fresh low-level server instances through resolveOriginalTool.
  (2026-10-05)

v0.21.4

Compare Source

Patch Changes

• #​5161 9e42cc2 Thanks @​pauldambra! - Make oversized events cheaper to truncate. The depth reduction now starts at the first depth that removes anything, so shallow payloads such as rows of data no longer normalize and measure the whole event again for each depth that changes nothing.
  (2026-10-02)

v0.21.3

Compare Source

Patch Changes

• #​5186 80b9e5f Thanks @​lucasheriques! - Sanitize only the part of a long string that the captured event can keep, so a tool call returning megabytes of HTML or text no longer blocks the event loop while it is scanned. A 10 MB HTML response now costs about 5 ms to capture instead of 2.7 s. The captured event does not change.
  (2026-10-02)

v0.21.2

Compare Source

Patch Changes

• #​5160 1ed05cf Thanks @​pauldambra! - Stop copying the full tool list into $mcp_response on $mcp_tools_list events, so instrumented servers answer tools/list faster.
  (2026-09-30)
• Updated dependencies [2d2560a]:
  • posthog-node@​5.55.0

v0.21.1

Compare Source

Patch Changes

• #​5157 e2a3e3d Thanks @​gesh! - Add the configured server build to custom events captured through PostHogMCP.
  (2026-09-30)

v0.21.0

Compare Source

Minor Changes

• #​5136 be66818 Thanks @​gesh! - Add optional MCP server build metadata
  (2026-09-28)

v0.20.0

Compare Source

Minor Changes

• #​5130 4d58499 Thanks @​gesh! - Record declared input aliases in automatic MCP instrumentation.
  (2026-09-28)

v0.19.0

Compare Source

Minor Changes

• #​5117 51699b4 Thanks @​pauldambra! - Record which declared parameter aliases a tool call relied on as $mcp_input_aliases_used, from a server-owned inputAliases map.
  (2026-09-28)

• #​5048 466da07 Thanks @​pauldambra! - Record safe tool input field names for automatic and custom MCP servers. Unknown names are [redacted] by default; shouldRecordInputKey replaces that rule.
  (2026-09-28)

v0.18.1

Compare Source

Patch Changes

• #​5112 74295fd Thanks @​gesh! - Keep MCP conversation sessions consistent across user messages.
  (2026-09-25)

v0.18.0

Compare Source

Minor Changes

• #​5074 7133bdb Thanks @​gesh! - Add conversation and session correlation helpers for custom MCP dispatchers.
  (2026-09-23)

v0.17.0

Compare Source

Minor Changes

• #​4924 be40430 Thanks @​lucasheriques! - Enable model capture and conversation correlation by default. Advertised tool schemas gain an llm_model argument (never enforced at dispatch) and eligible tool results gain a conversation handle; instrument(server, posthog, { captureModel: false, enableConversationId: false }) restores the previous shape. Fresh low-level instances now read both arguments under the ADR-0011 rule instead of staying silent.
  (2026-09-17)

v0.16.4

Compare Source

Patch Changes

• #​5009 b4f58eb Thanks @​gesh! - Use conversation IDs for get_more_tools and send_feedback calls handled by fresh server instances.
  (2026-09-17)

v0.16.3

Compare Source

Patch Changes

• #​4967 916e163 Thanks @​gesh! - Inject the get_more_tools tool only on the first tools/list page, so a paginated catalogue's concatenated listing carries it once instead of once per page. A real first-page tool with the same name still wins (warning logged); a real tool on a later page is shadowed, with a warning when a client fetches that page — rename the SDK's tool with the missingCapabilityToolName option if your catalogue uses the name.
  (2026-09-16)

v0.16.2

Compare Source

Patch Changes

• #​4953 2a6ddb9 Thanks @​gesh! - Inject the send_feedback tool only on the first tools/list page (the request with no cursor), so a paginated catalogue's concatenated listing carries it once instead of once per page. A real first-page tool with the same name still wins: the SDK warns, skips injection, and forwards its calls. A real tool that only appears on a later page is not detected — the SDK logs a warning when a client fetches that page; rename the SDK's tool with collectFeedback: { toolName } if your catalogue uses the name.
  (2026-09-15)

v0.16.1

Compare Source

Patch Changes

• #​4915 a5c1182 Thanks @​gesh! - Harden feedback validation, error handling, and tool-name collision routing.
  (2026-09-14)

v0.16.0

Compare Source

Minor Changes

• #​4870 cc6373b Thanks @​gesh! - Add the send_feedback virtual tool (new collectFeedback option): an honest, general agent-feedback channel with missing capabilities as the priority category. Every call emits a new $mcp_feedback event with $mcp_feedback_type and the other $mcp_feedback_* properties. Hosts can rename the tool, replace its description, declare extraProperties (captured as $mcp_feedback_<key>), and route reports to a real backend via onFeedback (instrument() path) or prepareToolCall().feedbackReport + captureFeedback() + sendFeedbackResult() (custom-dispatcher path). reportMissing / get_more_tools / $mcp_missing_capability are unchanged; new integrations should enable only collectFeedback.
  (2026-09-11)

v0.15.0

Compare Source

Minor Changes

• #​4830 39420d8 Thanks @​lucasheriques! - Capture resource discovery and reads from instrumented MCP servers. URL credential redaction (userinfo, credential-named query and fragment parameters) now applies to every captured string, including existing $mcp_tool_call parameters, responses, and error messages, so URLs already flowing through tool-call data will show %5Bredacted%5D values after upgrading.
  (2026-09-10)

v0.14.1

Compare Source

Patch Changes

• #​4798 4358915 Thanks @​posthog! - fix(error-tracking): collapse repeated frame cycles in parsed stack traces to reduce grouping differences caused by recursion depth, while preserving distinct throw locations
  (2026-09-09)
• Updated dependencies [4358915]:
  • @​posthog/core@1.51.1
  • posthog-node@​5.51.8

v0.14.0

Compare Source

Minor Changes

• #​4829 6724f10 Thanks @​lucasheriques! - Capture model identity from recognized client metadata before falling back to the injected llm_model argument.
  (2026-09-08)

v0.13.1

Compare Source

Patch Changes

• #​4783 6e9f931 Thanks @​gesh! - Reduce personal data in $mcp_intent, which is agent-narrated free text and could previously carry personal data a model read aloud.

  • Automatically redact structured personal identifiers — email addresses, phone numbers, IPv4/IPv6 addresses, Luhn-valid card numbers, and US SSNs — from $mcp_intent before it is captured. Redaction is always on, scoped to the intent only (structured tool arguments and results are untouched, since the same shapes are often legitimate data there), and best-effort for those well-defined shapes rather than free-form names or addresses. Identifiers grouped with Unicode spaces, dots, or slashes (as produced by copy-paste) and SSNs with space/dot separators are recognized. The email pattern uses bounded quantifiers so a pathological intent cannot cause quadratic-time backtracking.
  • Strengthen the default injected context prompt so agents are less likely to write personal data in the first place: the privacy rule is now explicit and lists the identifiers to avoid, and it tells the agent to refer to people and accounts by role ('a user', 'the customer') rather than by identity.

  context: false and beforeSend remain the ways to drop the field entirely. (2026-09-07)

• Updated dependencies [74ca945]:

  • posthog-node@​5.51.7
  • @​posthog/core@1.50.6

---

Configuration

📅 Schedule: (in timezone UTC)

• Branch creation
  • "before 6am on Monday"
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.