Niches · Agent observability & evals

Let a coding agent read traces over MCP or SQL

Two or three companies' docs show a full fit.

2companies, full fit
14companies, fit with a limit
24companies carry the tag
6share of the category's work
7first-hand accounts

What counts as done

The agent lists tools, discovers the schema and returns the matching traces or an aggregate, on a read-only credential, without the whole trace body in context.

Measured, for our own tests: it takes under 3 tool calls. No docs we read state these figures, so they don't count against a listing here.

How agents ask for it: ask Claude Code why errors spiked, query the traces with SQL, connect the MCP server read-only, fix this failing span from my editor

Does the job 2

The docs show it meets the test above, and nothing we read says otherwise.

ListingWhy
Overmind C
Overmind Technology Ltd (England and Wales, company 16585996)
MCP over OAuth or API keys with a read scope that exposes only inspection tools: query_traces, query_failures, inspect_capability_health and project resources, plus overmind://interface/current for the catalogue. source
Respan API + MCP BB
Keywords AI Inc. (d/b/a Respan)
A hosted MCP at mcp.respan.ai with a readonly endpoint and Read-scope keys, tool_search to find tools and log_list with filters; the CLI lists logs and traces and gets one by id. source

Does the job, with a limit 14

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

ListingWhy
agents-tree B
No legal entity found. The LICENSE and pyproject.toml name Karel Blavka
agents-tree --json emits a schema-versioned document of sessions and agents that an orchestrating agent can read; no MCP, SQL or credential scope, and prompts appear unredacted. source
Arize Phoenix BB
Arize AI, Inc.
Remote MCP at /mcp (beta) with search, get_schema and execute code-mode tools, OAuth or API key, and a viewer role for read-only accounts; px CLI prints JSON. Up to 50 call_tool invocations per execute and the endpoint is still labelled beta. source
Braintrust API + MCP B
Braintrust Data, Inc.
sql_query over project_logs with spans, traces or summary shapes, 10 rows by default and large results as signed URLs; resolve_object maps names to IDs. No read-only mode, all 50 tools load with the key's full permissions and hosted BTQL allows 20 queries a minute. source
Galileo API + MCP D
Galileo Technologies, Inc.
The MCP server mostly creates datasets and prompts; Get Signals is the only read of production data, with a Galileo-API-Key header and no read-only scope. Traces are read through the REST API instead. source
Helicone AI Gateway + MCP E
Helicone, Inc.
MCP query_requests and query_sessions take filters, time ranges and optional bodies; HQL is SQL with row-level security but only for selected workspaces. A use_ai_gateway tool can spend money and no read-only key scope is documented. source
HoneyHive D
HoneyHive Inc.
No trace MCP, only a docs MCP; a coding agent reads traces through the CLI with HH_PROJECT_API_KEY or the events search API on a read-only hh_ro_ key. No SQL. source
Laminar API + MCP C
LMNR AI, Inc.
Three MCP tools: ask_agent, query_laminar_sql (SELECT only, project-scoped) and get_trace_context, authenticated with a project API key over HTTP. No read-only key scope is documented and one user's MCP authentication kept failing (issue 1515). source
Langfuse API + MCP BB
Langfuse GmbH (a ClickHouse, Inc. company)
MCP server with listObservations, selectable fields and an isRootObservation filter on project-scoped keys, but 98 tools load read and write by default and read-only needs a client allowlist; one self-hoster found no trace read tools at all (issue 15142). source
LangSmith API + MCP BB
LangChain, Inc.
Remote MCP over OAuth or X-Api-Key with runs, traces, threads and datasets tools, scoped to the account's permissions; fetch_runs pages by a character budget. No read-only mode beyond roles, four write-named tools return documentation, and GET /runs/:id allows 30 calls a minute. source
LangWatch B
Reasoning Engine B.V. (doing business as LangWatch)
discover_schema then search_traces returns a model-sized digest, and Restricted keys limit a category to read. All 110 tools load by default, deletes and key creation included, with no annotations. source
MLflow Tracing C
MLflow Project, a Series of LF Projects, LLC
The experimental MCP server has search_traces with filter_string status='ERROR' and extract_fields to keep bodies out of context, and MLFLOW_MCP_TOOLS=traces trims the set. Read-only needs an RBAC user; delete_traces runs without confirmation. source
Opik API + MCP BB
Comet ML, Inc.
Five MCP tools: read returns a trace with its span tree, list filters by status with a time window, schema gives the filter vocabulary, nothing can be deleted. No read-only mode and the key carries the user's full rights; listing is capped at 30 requests a minute. source
Pydantic Logfire B
Pydantic Services Inc.
Remote MCP with a project:read key or OAuth, query_schema_reference before query_run SQL, and an Enterprise-only org-wide read-only policy; queries draw on an unpublished daily budget and return 429. The Fix AI command errored for one user (issue 1485) and another wants psql (1451). source
W&B Weave B
Weights and Biases, LLC
The Weights & Biases MCP server at mcp.withwandb.com answers questions about traces and evaluations with a Forge API key; no read-only scope or trace-summary tool is documented on the page read. source

Doesn't do this job 6

Listings in the category that do something else, with the reason.

ListingWhy
Agent Tree C
No legal entity found. The LICENSE, plugin.json and marketplace.json name "Mike Pollow, Digitalroots"
No MCP, SQL or JSON output. source
BotGauge E
BotGauge AI Inc. (Delaware, United States)
No MCP or SQL; the REST API covers evaluators and rules, not trace reads. source
DeepEval B
Confident AI, Inc.
No MCP or SQL read path over traces. source
Gauge CLI B
Gauge
The API reads eval results with a read-only key, not traces. source
Navige E
Navige Ltd
The MCP route governs tool servers; there is no read surface over recorded calls for a coding agent. source
Prefactor B
Prefactor Pty Ltd
No MCP or SQL; a read_only token can list spans over REST, but there is no schema discovery or summary tool for a coding agent. source

What stops agents finishing

  • MCP server with no read-only mode
  • MCP exposes no trace read tools
  • MCP auth fails for self-hosted
  • query budget per day with no published number

Around it

Before: agent-harnesses.pair-in-terminal. After: agent-observability.debug-failed-run.

Where this comes from

Weight 6 of 100 in Agent observability & evals: 7 of 86 accounts: Logfire issues on psql and the Fix AI command, the Tiger Data post on Logfire MCP from Claude Code, Langfuse's MCP without trace reads, Laminar MCP auth, the Phoenix export request, the sub-agent cost CLI

The same as data: /api/v1/use-cases/agent-observability.agent-queries-traces.json, or the MCP tool get_use_case with id agent-observability.agent-queries-traces. Other jobs in this category: /api/v1/niches?category=agent-observability.

Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person. A mark says what the docs show, not what a run measured.

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.