feat: adds MCP metrics to OTEL

[roroghost17]

Summary

MCP operations were missing OTel tracing spans and the mcp.client.operation.duration metric because the tracer was never stamped onto request-scoped contexts for standalone MCP tool executions, and the plugin pipeline emitted a single flat span rather than following the nested span layout used by the LLM path.

Changes

• Added ensureMCPTracerContext to stamp the tracer onto the BifrostContext for ExecuteChatMCPTool and ExecuteResponsesMCPTool calls, which bypass the chat/responses flow that normally sets it.
• Restructured span creation in RunWithPluginPipeline and runConnectWithPluginPipeline to match the LLM path: PreHook spans chain under the inbound HTTP span, the op span is created after PreHooks complete (nesting under the last one), and PostHook spans nest under the op span. Short-circuited requests produce no op span.
• Added SpanKindMCPTool for tools/call spans, distinct from SpanKindMCPClient used for lifecycle ops (ping, list_tools, connect).
• Added mcp.method.name (tools/call, tools/list, ping) and network.transport (pipe for stdio, tcp for HTTP/SSE) OTel semconv attributes on MCP spans.
• Added bifrost.mcp.tool.duration_ms to carry the raw CallTool wire latency so the duration metric measures the wire round-trip rather than span wall-time (which includes PostHooks).
• Added error.type classification on failed MCP spans via sentinel errors (ErrMCPToolTimeout, ErrMCPToolCallFailed) wrapped with %w at the CallTool site, classified by mcpErrorType via errors.Is.
• Added setMCPGovernanceSpanAttrs to copy governance identity (bifrost.virtual_key_id, team_id, etc.) from context onto MCP spans for per-tenant metric breakdowns.
• Added mcp.client.operation.duration histogram in the OTel metrics exporter with semconv-recommended bucket boundaries. recordMCPMetricsFromTrace emits one sample per MCP span, preferring wire latency over wall-time and appending error.type on failures.
• Preserved the original error value on ErrorField.Error (json:"-") when wrapping opErr as BifrostError so errors.Is chains work through PostHook boundaries.
• Added OTelMethodName() on MCPRequestType and OTelNetworkTransport() on MCPConnectionType for semconv mapping.
• Added connectionTypeForClientName to resolve a client's transport by name for the network.transport attribute.

Type of change

• Bug fix
• Feature
• Refactor
• Documentation
• Chore/CI

Affected areas

• Core (Go)
• Transports (HTTP)
• Providers/Integrations
• Plugins
• UI (React)
• Docs

How to test

go test ./core/mcp/... ./plugins/otel/...

TestMCPErrorTypeClassification validates sentinel-based error.type classification including auth_required, timeout, tool_error, and the _OTHER catch-all.

TestRecordMCPMetricsFromTraceRecordsBothKinds validates that mcp.client.operation.duration emits one data point per enriched MCP span (SpanKindMCPTool and SpanKindMCPClient), uses wire latency over wall-time when available, tags error.type on failures, and skips un-enriched and non-MCP spans.

TestBuildMCPSpanAttrsSemconvAndGovernance validates that semconv dimensions and flat-named governance labels are present and that absent optional dimensions are omitted.

TestMCPRequestTypeOTelMethodName and TestMCPConnectionTypeOTelNetworkTransport pin the semconv mapping tables.

Breaking changes

• Yes
• No

Related issues

Security considerations

None. Governance identity values copied to span attributes are already present in the request context and are not new data exposures.

Checklist

• I read docs/contributing/README.md and followed the guidelines
• I added/updated tests where appropriate
• I updated documentation where needed
• I verified builds succeed (Go and UI)
• I verified the CI pipeline passes locally if applicable

[coderabbitai]

Warning

Review limit reached

You’ve reached a temporary PR review limit under our Fair Usage Limits Policy.

Your recent review volume is higher than typical usage, so adaptive limits are currently applied.

Next review available in: 51 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: cc6f82ce-8d8a-40d7-b255-f8a7bc081542

📥 Commits

Reviewing files that changed from the base of the PR and between 66bfb1c and 1f83507.

📒 Files selected for processing (10)

• core/bifrost.go
• core/mcp/error_classify_test.go
• core/mcp/pluginpipeline.go
• core/mcp/toolmanager.go
• core/schemas/bifrost.go
• core/schemas/mcp.go
• core/schemas/trace.go
• plugins/otel/main.go
• plugins/otel/mcp_metrics_test.go
• plugins/otel/metrics.go

📝 Walkthrough

Walkthrough

MCP execution now attaches missing tracers, classifies tool failures with sentinel errors, enriches MCP spans with standardized attributes, and exports MCP operation duration metrics with governance and error dimensions.

Changes

MCP observability

Layer / File(s)	Summary
MCP observability contracts core/mcp/pluginpipeline.go, core/mcp/toolmanager.go, core/schemas/...	Adds typed MCP tool errors, OpenTelemetry method and transport mappings, and MCP span attribute constants.
MCP pipeline span lifecycle core/mcp/pluginpipeline.go, core/mcp/error_classify_test.go	Moves operation spans after pre-hooks, preserves wrapped errors, applies governance and result attributes, and classifies failures.
Standalone MCP tracer context core/bifrost.go	Attaches a tracer to standalone chat and responses MCP execution contexts when absent.
MCP duration metrics plugins/otel/...	Defines and records mcp.client.operation.duration metrics from eligible MCP spans, including optional tool, transport, governance, and error attributes, with coverage tests.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant MCPEntryPoint
  participant MCPManager
  participant MCPTool
  participant OtelPlugin
  participant MetricsExporter
  MCPEntryPoint->>MCPManager: execute MCP request with tracer context
  MCPManager->>MCPTool: run tool through plugin pipeline
  MCPTool-->>MCPManager: result or classified error
  MCPManager->>OtelPlugin: complete enriched MCP span
  OtelPlugin->>MetricsExporter: record operation duration and attributes

Loading

Suggested reviewers: akshaydeo, pratham-mishra04

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is concise and accurately summarizes the PR’s main MCP OpenTelemetry work.
Description check	✅ Passed	The description follows the template and includes summary, changes, testing, and checklist details.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch 07-15-feat_adds_mcp_metrics_to_otel

---

_{Comment @coderabbitai help to get the list of available commands.}

[CLAassistant]

All committers have signed the CLA.

[roroghost17]

• feat: adds user email export to connectors #5302
• docs: connector docs update for MCP metrics #5239
• feat: add mcp metrics to telemetry plugin #5238
• feat: adds MCP metrics to OTEL #5237 👈 (View in Graphite)
• dev

This stack of pull requests is managed by Graphite. Learn more about stacking.

[greptile-apps]

Confidence Score: 5/5

This looks safe to merge.

• The inbound HTTP path creates and flushes the trace in tracing middleware.
• The updated executor supplies the tracer needed to create MCP child spans.
• No blocking issue remains in the changed code.

Important Files Changed

Filename	Overview
core/bifrost.go	Propagates the configured tracer to MCP tool execution contexts that already carry a transport trace.
core/mcp/pluginpipeline.go	Restructures MCP spans around plugin hooks and adds transport, governance, latency, and error attributes.
core/mcp/toolmanager.go	Wraps MCP tool timeout and call failures with classification sentinels.
plugins/otel/main.go	Records MCP operation duration metrics from enriched spans in completed traces.
plugins/otel/metrics.go	Registers the MCP client operation duration histogram and recommended buckets.

_{Reviews (3): Last reviewed commit: "feat: adds MCP metrics to OTEL" | Re-trigger Greptile}

[coderabbitai]

Actionable comments posted: 2

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

core/bifrost.go (1)

2826-2831: 🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift

Keep operation-specific span state out of reusable contexts.

The standalone entrypoints can pass the shared bifrost.ctx, while both MCP span starters overwrite its current span ID. Concurrent or subsequent calls can consequently receive stale or cross-request parent spans.

• core/bifrost.go#L2826-L2831: create a request-scoped context instead of using bifrost.ctx directly.
• core/bifrost.go#L2845-L2850: apply the same isolation to the Responses entrypoint.
• core/mcp/pluginpipeline.go#L88-L93: propagate the operation span through scoped state rather than permanently mutating the caller context.
• core/mcp/pluginpipeline.go#L302-L307: apply the same fix to connection spans.

Based on learnings, BifrostContext.SetValue mutates the shared context in place.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@core/bifrost.go` around lines 2826 - 2831, Isolate operation-specific MCP
span state from reusable contexts. In core/bifrost.go lines 2826-2831, update
ExecuteChatMCPTool to derive a request-scoped context instead of assigning
bifrost.ctx directly; apply the same isolation to the Responses entrypoint at
lines 2845-2850. In core/mcp/pluginpipeline.go lines 88-93 and 302-307,
propagate span state through scoped context values without mutating the caller’s
shared BifrostContext.

Sources: Path instructions, Learnings

🧹 Nitpick comments (1)

core/mcp/toolmanager.go (1)

713-718: 🗄️ Data Integrity & Integration | 🔵 Trivial | ⚡ Quick win

Preserve the underlying callErr in the error chain.

Using %v flattens the client error, so callers can classify the sentinel but cannot use errors.Is/errors.As on the original cause. Wrap both errors, for example with Go’s multiple %w support.

As per coding guidelines, Go errors should be explicitly wrapped.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@core/mcp/toolmanager.go` around lines 713 - 718, Update the MCP tool-call
failure return in the surrounding tool execution method to preserve both callErr
and ErrMCPToolCallFailed in the error chain by using multiple %w wrapping. Keep
the existing timeout handling and error message context unchanged.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@core/mcp/pluginpipeline.go`:
- Around line 316-322: Update the deferred span-finalization logic around
setMCPGovernanceSpanAttrs and tracer.EndSpan to track operation and post-hook
failures, ending failed connection spans with an error status and the relevant
error details instead of always using SpanStatusOk. Preserve successful
operations as SpanStatusOk and retain the existing nil guards.
- Around line 110-119: Update the tool argument and result attribute capture in
the response handling flow around GetToolArguments and the corresponding result
logic so content is not exported by default. Require an explicit opt-in before
setting these attributes, redact sensitive fields, and truncate or reject values
exceeding the configured size limit while preserving existing nil handling.

---

Outside diff comments:
In `@core/bifrost.go`:
- Around line 2826-2831: Isolate operation-specific MCP span state from reusable
contexts. In core/bifrost.go lines 2826-2831, update ExecuteChatMCPTool to
derive a request-scoped context instead of assigning bifrost.ctx directly; apply
the same isolation to the Responses entrypoint at lines 2845-2850. In
core/mcp/pluginpipeline.go lines 88-93 and 302-307, propagate span state through
scoped context values without mutating the caller’s shared BifrostContext.

---

Nitpick comments:
In `@core/mcp/toolmanager.go`:
- Around line 713-718: Update the MCP tool-call failure return in the
surrounding tool execution method to preserve both callErr and
ErrMCPToolCallFailed in the error chain by using multiple %w wrapping. Keep the
existing timeout handling and error message context unchanged.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ce9ae58e-b73e-4c8a-a658-0acf404a802e

📥 Commits

Reviewing files that changed from the base of the PR and between c0909f9 and efa1e88.

📒 Files selected for processing (10)

• core/bifrost.go
• core/mcp/error_classify_test.go
• core/mcp/pluginpipeline.go
• core/mcp/toolmanager.go
• core/schemas/bifrost.go
• core/schemas/mcp.go
• core/schemas/trace.go
• plugins/otel/main.go
• plugins/otel/mcp_metrics_test.go
• plugins/otel/metrics.go

[Pratham-Mishra04]

Merge activity

• Jul 17, 11:11 AM UTC: A user started a stack merge that includes this pull request via Graphite.
• Jul 17, 11:12 AM UTC: Graphite rebased this pull request as part of a merge.
• Jul 17, 11:14 AM UTC: Graphite rebased this pull request as part of a merge.
• Jul 17, 11:15 AM UTC: @Pratham-Mishra04 merged this pull request with Graphite.