How is this different from connecting Braintrust MCP directly?

Connected directly, an agent sees every tool the server exposes and holds the credential itself. Through TrustGate you choose which tools each agent gets, and every call is logged against the agent that made it.

What if the Braintrust MCP API changes?

We track the Braintrust MCP schema, so the tool catalogue stays current. Tools you have already allowed keep working, and anything new stays denied until you enable it.

Can I give different agents different Braintrust tools?

Yes. Access is set per agent: one can get a handful of read tools, another the full set. You grant the whole MCP or single tools within it.

Where can I see what my agents did in Braintrust?

Every call is logged against the agent that made it, with the tool it used. Activity can be sent on to your own monitoring.

Do I need to run the Braintrust MCP server myself?

No. TrustGate connects to the Braintrust MCP server and sits in front of it. Your agent only talks to your TrustGate URL, never to Braintrust directly.

Connect Braintrust MCP to your AI agents, securely.

Q: Is Braintrust MCP secure?

The server itself exposes 47 tools, and some of them can write to your Braintrust account. Behind TrustGate an agent only sees the tools you allow, requests and responses are inspected, and every call is logged. Anything outside the allowlist is denied before it reaches Braintrust.

One secured endpoint for every Braintrust call your AI agents make, scoped per user, inspected, and logged.

Get started for free

Get a demo

Try Braintrust MCP now

What your agents can ask once Braintrust MCP is connected.

Select the tools your AI agents can call

47 tools from the Braintrust MCP schema.

Showing 47 tools

sql_queryreadQueries experiments, datasets and logs using SQL, supporting SELECT, FROM, WHERE, GROUP BY, ORDER BY and LIMIT. Results over 1 MB are uploaded to object storage and returned as a signed URL instead of inline rows.
infer_schemareadDiscovers the available fields, data types and most common values in experiments, datasets or logs.
summarize_experimentreadReturns aggregated performance metrics for an experiment, optionally compared to a baseline.
list_recent_objectsreadLists recently created projects, experiments, datasets, prompts or functions the account can access.
resolve_objectreadConverts names to IDs or the reverse, and parses Braintrust URLs. Ambiguous names return an error.
generate_permalinkreadGenerates a direct web link to a Braintrust object for sharing or bookmarking.
lookup_usersreadResolves organization member IDs to names and emails, searches by partial name or email, or lists every member.
lookup_api_keysreadResolves API key and service token IDs to their names, following existing key and token visibility rules.
get_trace_work_itemsreadReturns a chronological list of the meaningful LLM and tool spans in a trace.
update_trace_work_reportwriteSaves work sections and per-span annotations back onto the trace.
search_patternsreadSearches project patterns by text, semantic similarity, ID, status or quality, to catch duplicates that use different wording.
new_patternwriteRecords a pattern the agent has identified, with the evidence behind it. The agent decides what counts as a pattern.
update_patternwriteUpdates an existing project pattern, or attaches evidence to matching traces.
test_preprocessor_on_tracereadRuns a saved, global or inline preprocessor on up to 50 span, trace or group references without writing to the source trace.
test_facet_on_tracereadRuns an inline facet definition on up to ten span, trace or group references without writing the result back.
create_preprocessorwriteCreates a versioned preprocessor from inline JavaScript that converts raw trace data into text.
create_facetwriteCreates a versioned facet that extracts a short summary from spans or traces, using Braintrust's built-in facet model.
enable_topics_automationhigh riskEnables Topics for a project, seeding processing for new traffic. It does not rewind historical data.
set_topics_automationhigh riskUpdates an existing Topics automation's facets, scope, filters, sampling or timing.
rewind_topics_automationhigh riskRewinds a Topics automation to process historical data from a start time or recent window. This draws from the account's monthly model credits.
generate_monitor_chartreadPreviews a monitor chart against real project logs without modifying a saved view.
list_monitoring_viewsreadLists a project's saved monitor views and their chart IDs.
get_monitoring_viewreadInspects a saved monitor view, including its options and ordered chart definitions.
create_monitoring_viewwriteCreates a project-scoped monitor view, optionally containing charts already previewed.
update_monitoring_viewwriteInserts, updates or removes charts in an existing monitor view, one edit at a time or in bulk.
list_automationsreadLists a project's automations including online scoring rules, alerts, exports, retention policies and Topics automations, with their complete configurations.
list_slack_channelsreadLists connected Slack workspaces and public channels for a project, to resolve IDs before configuring Slack delivery.
create_log_alertwriteCreates an alert for individual matching project logs, with an interval to throttle repeated notifications.
create_environment_update_alertwriteCreates an alert for environment updates, optionally limited to specific environment slugs.
create_threshold_alertwriteCreates an alert on an aggregate over a recent window of project data, evaluated on a schedule: averages, counts, rates, percentages, percentiles and distributions.
set_automation_statushigh riskPauses or activates an alert, scheduled job or online scoring rule. Pausing silences monitoring without deleting anything.
create_scheduled_loop_jobhigh riskCreates a Loop job that runs on an interval or cron schedule over a recent window of project data, unattended afterwards.
test_evaluatorreadRuns a saved, global or inline evaluator against span, trace or group references without writing results to the source trace.
create_promptwriteCreates a versioned prompt from a completion-style prompt or chat messages, optionally replacing an existing one as a new version.
create_evaluatorwriteCreates a versioned LLM or inline code evaluator, producing numeric scores or categorical labels.
create_human_review_scorewriteCreates a categorical, slider or free-form human review score for a project. It fails if the name already exists.
update_online_scoring_rulehigh riskSaves or rewinds an online scoring rule that runs saved evaluator functions against production logs. New rules default to paused.
add_to_datasetwriteCopies project log rows into a dataset by row ID, preserving input, metadata and tags with an origin link to the source log. Up to 1,000 rows per call on hosted Braintrust.
run_evalhigh riskRuns an experiment with a hosted dataset, inline rows or a prior experiment as input, and any saved or inline task and scorers. It can execute your code or call AI providers, so it incurs compute and model usage.
edit_dataset_rowshigh riskInserts, updates or deletes dataset rows, and can create a new named dataset. It can permanently delete rows.
save_dataset_custom_viewwriteSave a [custom view](/docs/annotate/custom-views.md) for labeling a dataset's rows and return a link that opens the dataset with the view selected.
list_dataset_custom_viewsreadList up to 100 of a project's saved dataset custom views, optionally for one dataset, with each view's code and a link that opens it.
add_to_human_review_queuewriteAdd up to 100 project log traces, experiment traces, or dataset rows from one source to [human review](/docs/annotate/human-review/score-traces.md) by row ID.
get_project_settingsreadReturns a project's typed settings, including the effective default preprocessor.
search_docsreadSearches Braintrust documentation for guides, API references and code examples.
load_braintrust_skillreadLoads a Braintrust workflow guide before using the tools it covers: automations-workflow, evaluator-workflow or topics-workflow. Loading is read-only and costs one tool call.
set_project_default_preprocessorwriteSets or clears a project's default preprocessor, which changes what facets and other project functions inherit when they do not select one explicitly.

What Braintrust MCP exposes

Braintrust MCP groups its 47 tools into 10 categories, each with a suggested starting policy.

All 47 tools as JSON, ready to paste.

Category	Covers	Access	Suggested policy
Topics	Versioned preprocessors that turn raw trace data into text, facets that extract short summaries, testing both against real traces without writing back, and enabling, updating and rewinding the Topics automation	Read / Write	rewind_topics_automation reprocesses historical traffic and draws down monthly model credits. The test tools are the safe path and Braintrust expects the topics-workflow skill loaded first so each stage is validated before saving
Automations & Alerts	Listing a project's automations including online scoring rules, exports, retention policies and Topics automations with their full configurations, pausing and activating them, creating log, environment-update and threshold alerts, scheduled Loop jobs, and resolving Slack workspaces and channels	Read / Write	set_automation_status can pause an alert, which silences monitoring without deleting anything. create_scheduled_loop_job sets work running on a cron afterwards
Objects & Identity	Listing recent projects, experiments, datasets, prompts and functions, converting names to IDs and parsing Braintrust URLs, generating shareable permalinks, and resolving member and API key IDs to names and emails	Read	Read-only. lookup_users returns member names and emails and lookup_api_keys resolves key and service token IDs to names, following existing visibility rules. generate_permalink produces links meant to be shared outside the conversation
Dashboards	Previewing a monitor chart against real project logs without saving, listing and inspecting saved monitor views with their chart definitions, and creating or editing views one chart at a time or in bulk	Read / Write	generate_monitor_chart previews without touching a saved view, so it is the right first step. update_monitoring_view can remove charts as well as add them
Prompts & Evaluators	Creating versioned prompts, LLM or inline code evaluators and human review scores, testing an evaluator against traces without writing results back, and saving or rewinding an online scoring rule	Read / Write	update_online_scoring_rule puts an evaluator against production logs continuously. New rules default to paused, so activation is the moment to approve, not creation
Settings & Docs	Reading a project's typed settings including the effective default preprocessor, setting or clearing that default, searching Braintrust documentation, and loading a workflow skill	Read / Write	set_project_default_preprocessor changes what facets and other project functions inherit, so it propagates beyond the one object it names
Data & Query	SQL over experiments, datasets and logs, discovering the fields, types and most common values in a data source, and aggregated experiment metrics with an optional baseline comparison	Read	Read-only, but sql_query reaches production LLM traces, which contain whatever users and the application sent. Results over 1 MB leave as a signed URL to object storage rather than inline rows
Patterns	Searching project patterns by text, semantic similarity, ID, status or quality to catch duplicates worded differently, recording a new pattern with its evidence, and updating one or attaching further evidence	Read / Write	new_pattern writes the agent's own conclusions into the project record, and Braintrust notes the agent decides what counts, so asking for one does not guarantee a record
Evals & Datasets	Running an experiment against a hosted dataset, inline rows or a prior experiment with saved or inline tasks and scorers, inserting, updating and deleting dataset rows, and copying project log rows into a dataset by row ID	Read / Write	run_eval executes code and calls AI providers, so it costs money per call. edit_dataset_rows can permanently delete rows; both it and add_to_dataset cap at 1,000 rows per call on hosted Braintrust
Traces	Returning the meaningful LLM and tool spans in a trace in chronological order, and saving work sections and per-span annotations back onto that trace	Read / Write	update_trace_work_report writes the agent's annotations onto a real trace, so the analysis becomes part of the record other people read

Connect Braintrust MCP to your AI agent

Four steps, under ten minutes.

01
Sign in to NeuralTrust
Open the NeuralTrust platform, sign in and go to Getting started, then Connect your tools.
02
Choose your agent
When asked which agent you use, pick Claude Code, Codex, Cursor, Windsurf, Copilot or your own code.
03
Connect Braintrust
Find Braintrust in the tool list and connect it.
04
Connect your agent
Follow the steps on screen to add the config to your agent, then send the test prompt.

Permissions down to the single tool

You do not allow Braintrust MCP or block it. You allow the exact tools each agent can call, one by one.

Braintrust MCP	47 TOOLS AVAILABLE
Code review agent	4 tools, read only
Issue triage bot	9 tools
Platform team assistant	all 47 tools

Tool by toolConnecting the server is not the unit of access. Each tool is granted on its own.
Default denyA tool you have not enabled is never exposed to the agent.
Approval on writesTools that change or delete things can require a human to confirm.

Is Braintrust MCP secure?

One endpoint instead of raw credentials. Every call is checked against policy and logged.

AI agentsConnect through your MCP client configuration.	NeuralTrustApply controls based on your configuration.	Braintrust MCP47 documented tools over streamable HTTP; SSE-only clients cannot connect. OAuth with dynamic client registration or an API key, acting with that account's permissions. Request bodies cap at 100 MiB, and three workflow skills are loaded through the server itself.
Coding assistants	Connection & credentialsper user	Topics7 tools
Internal copilots	Tool access policiesdeny by default	Automations & Alerts7 tools
Workflow automations	Request & response inspectioninline	Evals & Datasets6 tools
Compatible MCP clients	Activity monitoringalerts and logs	All other tools27 tools