Skip to main content
Glama
log-10x

Log10x MCP Server

Official
by log-10x

POC from SIEM (submit)

log10x_poc_from_siem_submit

Start a log cost optimization POC against your SIEM: pulls a sample, templatizes patterns, and generates a report with cost drivers and native exclusion configs.

Instructions

Kick off a full log-cost-optimization POC against the user's SIEM. Pulls a representative event sample, templatizes into stable pattern identities, and renders a 9-section markdown report covering top cost drivers, Receiver recommendations, ready-to-paste native SIEM exclusion configs, Compact mode potential, risk/dependency checks, and deployment paths. Supported SIEMs: cloudwatch (AWS CloudWatch Logs via IAM credential chain), datadog (DD_API_KEY + DD_APP_KEY), sumo (Sumo Logic), gcp-logging (GCP Cloud Logging), elasticsearch (Elastic Cloud / self-hosted), azure-monitor (Azure Monitor / Log Analytics), splunk (SPLUNK_HOST + SPLUNK_TOKEN), clickhouse (OpenObserve / SigNoz / custom schemas). Auto-detects the SIEM from env vars when siem omitted — explicitly pass siem if multiple credential sets exist. scope and query are SIEM-specific: CloudWatch (log group + filter pattern), Datadog (index + query), Sumo (_sourceCategory + query), GCP (project id + filter), Elasticsearch (index pattern + KQL), Azure (workspace id + KQL), Splunk (index + SPL), ClickHouse (database + SQL WHERE). For ClickHouse, also pass clickhouse_table (required) and column-mapping args for custom schemas (OpenObserve/SigNoz auto-detected). Returns a snapshot_id — poll via log10x_poc_from_siem_status to retrieve progress and the final report. Report is also written to ${LOG10X_REPORT_DIR:-/tmp/log10x-reports}/poc_from_siem-<timestamp>.md. Default window is 7d, default target event count is 250k, default max pull time is 5 min — the pull stops at whichever of the two ceilings hits first. Tier prerequisites: none. No log10x API key required. Templating defaults to privacy_mode=true: events are templated by a local Log10x engine, either a native tenx CLI (install for macOS/Linux/Windows: https://doc.log10x.com/install/) or a local Docker container (set LOG10X_TENX_MODE=docker, auto-detected and preferred when the mode is unset), and never leave the machine. Set privacy_mode: false to route through the public Log10x paste endpoint — demo use only, not production log content. Two modes, and the agent should name whichever the user's phrasing invites. This tool is the NETWORKED one: it reads the analyzer directly over the network with the caller's credentials, and the MCP mints one licence. The FENCED mode runs the same POC with no network at all — log10x_emit_sample_plan renders a read-only export script the user runs themselves, and the analysis happens in a container started with --network none, so 'cannot exfiltrate' is a kernel fact the user checks with one docker inspect rather than a claim we make. Offer it; do not ask which one first.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
siemNoWhich SIEM to pull from. Omit to auto-detect from ambient credentials. Valid values: cloudwatch, datadog, sumo, gcp-logging, elasticsearch, azure-monitor, splunk, clickhouse.
queryNoSIEM-native filter expression layered on top of `scope`. Syntax per SIEM (CloudWatch filter pattern; Datadog query; KQL for ES/Azure; SPL for Splunk; SQL WHERE for ClickHouse; Sumo query).
scopeNoSIEM-specific resource scope. CloudWatch: log group name or wildcard (`/aws/ecs/*`). Datadog: index name. Sumo: `_sourceCategory`. GCP: project id. Elasticsearch: index pattern. Azure Monitor: workspace id. Splunk: index name. ClickHouse: database name.
windowNoWindow to pull over. Accepts "1h", "24h", "7d", "14d", "30d". Default "14d" — wide windows unlock the differentiated longitudinal signals (first-seen, growth, stable-vs-new) that the agent cannot compute from a small sample. Pull pacing is automatic; long windows take minutes but the snapshot continues in the background.14d
ai_prettifyNoDefault true: use MCP sampling to ask the host LLM (the same model the user is already chatting with — Claude Desktop, Claude Code, Cursor, etc.) to batch-generate 3-5-word human-readable names for the top patterns. No Log10x-side endpoint, no extra API key — the host uses whatever model + credentials the user already has. Sends only templated pattern identities (no variable values, no raw log content). Skipped automatically when the host does not advertise the `sampling` capability; the report falls back to raw snake_case identities plus a note. Set false to skip unconditionally.
environmentNoOptional environment nickname — cosmetic only, for the report header.
pin_patternsNoAdvanced — most customers will not need this. Map of pattern_hash to action for rare per-pattern overrides within a service. Applied AFTER pin_services. Use only when a single pattern inside an otherwise-reducible service must be excepted (e.g., audit-trail log line inside a chatty service).
pin_servicesNoPrimary per-service override surface. Map of service name to action. "Pin payment-svc to pass" → {"payment-svc":"pass"}. Pins are applied AFTER the destination default and AFTER exception_services. Feasibility math reruns with the pins; max_achievable_percent may shift and reason cites the pins.
total_daily_gbNoCustomer's total daily log volume in GB/day. Pick any one of total_daily_gb / total_monthly_gb / total_annual_gb — whichever unit the user naturally thinks in. The tool normalizes to daily internally. When any is provided (or auto_detect_volume succeeds), per-pattern costs are extrapolated from the pulled sample to the full volume, producing meaningful annual-savings figures instead of sub-cent numbers. Priority: daily > monthly > annual. If the pull was narrowed via `query` to one service, this overstates cost — only a fraction of daily volume matches the filter.
total_annual_gbNoCustomer's total annual log volume in GB/year. See total_daily_gb for semantics.
clickhouse_tableNo[ClickHouse] Required — table name holding log events.
max_pull_minutesNoHard cap on pull wall-time. Default 30. The pull stops at whichever of target_event_count, max_pull_minutes, or saturation-detected hits first. Long pulls run in the background; poll status while the user does other things.
total_monthly_gbNoCustomer's total monthly log volume in GB/month. See total_daily_gb for semantics.
enrich_max_tokensNoOutput token cap for the host-agent enrichment call. Default 8000.
auto_detect_volumeNoDefault true: when no total_*_gb arg is provided, probe the SIEM's usage/metrics API to auto-detect daily ingest volume. Per-SIEM best-effort: CloudWatch (describeLogGroups ÷ retention), Datadog (Usage API), Elasticsearch (_stats), Azure (Usage KQL table), GCP (Cloud Monitoring byte_count), ClickHouse (system.parts), Splunk (license API), Sumo (Account Usage API). Fails silently and falls back to scenario brackets if the current creds lack the required scope. Set false to skip the probe and go straight to manual args or scenarios.
exception_servicesNoServices flagged by the customer to stay in the SIEM with full retention (action=pass). Typically 1-3 services for audit, compliance, or executive dashboards. Patterns whose service is in this list are pinned to pass on the envelope outputs and their bytes are subtracted from the achievable reduction pool used for the feasibility verdict.
target_event_countNoTarget event count for the pull. Default 1,000,000 (~500 MB at 500B avg, tokenizes in 5-10 min). The pull self-terminates earlier on saturation — when new patterns per 100k events drops below 2%, the long tail has been covered and the report is generated. This default is intentionally two orders of magnitude beyond what an unaided agent can fit in context.
analyzer_cost_per_gbNoOverride the $/GB rate for cost calculations. Default is read from vendors.json per detected SIEM.
enrich_with_host_agentNoDefault true: after the engine produces measured findings (per-pattern $/mo, growth, incident clusters), ask the MCP host LLM via sampling to contribute operational context the engine cannot see: kubectl events / deploys correlating with GROWING patterns, alert / dashboard dependencies before recommending mute, code-level root-cause refinement on code_fix patterns, and prioritization based on customer context. Single round-trip, capped at 8000 output tokens. Skipped automatically when the host does not advertise sampling; the v2 envelope still ships without enrichment. Contributions land in output.agent_enrichment.contributions with an audit trail (tools_inspected) so the customer sees what the agent says it looked at.
target_percent_reductionNoCustomer-specified target reduction percent. If absent, POC produces a recommendation-only output. If present, POC produces a feasibility verdict (`output.feasibility`) plus a pre-deploy commitment artifact stub (`output.commitment_artifact`) the agent can surface alongside the per-pattern actions. The cap CSV ready to commit is attached by a later change.
clickhouse_message_columnNo[ClickHouse] Column holding the message body. Default auto-detected.
clickhouse_service_columnNo[ClickHouse] Optional column for service name.
clickhouse_severity_columnNo[ClickHouse] Optional column for severity.
clickhouse_timestamp_columnNo[ClickHouse] Column holding the timestamp. Default auto-detected.

Output Schema

TableJSON Schema
NameRequiredDescriptionDefault
dataNo
toolYes
viewNosummary
imagesNo
actionsNo
summaryYes
warningsNo
truncatedNo
next_cursorNo
render_hintNo
generated_atYes
schema_epochYes
schema_versionYes

Schema Changelog

Changes observed during successful MCP inspections.

  1. First observedv1.30.7

TDQS

A4.3/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Far exceeds the annotations (readOnlyHint=false, openWorldHint=true, idempotentHint=false). It discloses network behavior, a single licence minted, privacy_mode defaults with data residency, background execution, auto-detection failure fallback, and the FENCED-mode contrast — behavior an agent cannot infer from structured fields.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Purpose is front-loaded, but the body is very long and padded with persuasive wording (the 'kernel fact' exfiltration framing, 'Offer it; do not ask which one first') that goes beyond invocation guidance. Some length is justified by 24 parameters, yet much of the text duplicates the schema.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

An output schema exists, so return values need not be explained, and the description correctly points to snapshot_id polling. It covers the essential operational context (creds, auto-detect, background pulls), though the stale default values undermine completeness for callers trusting the prose over the schema.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters2/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the schema already documents all 24 parameters richly. Worse, the description contradicts the schema defaults: it claims 'Default window is 7d, default target event count is 250k, default max pull time is 5 min,' while the schema declares 14d, 1,000,000, and 30 minutes respectively. Misleading default values are worse than silence.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb+resource ('Kick off a full log-cost-optimization POC against the user's SIEM') and enumerates exactly what it does: pulls a sample, templatizes, renders a 9-section report. It also distinguishes itself from siblings, naming log10x_poc_from_siem_status for polling and log10x_emit_sample_plan for the fenced alternative.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly frames when to use this tool ('the NETWORKED one') versus the fenced mode that 'runs the same POC with no network at all,' and directs the agent to 'offer it; do not ask which one first.' It also names the exact follow-up tool (log10x_poc_from_siem_status) for retrieving results.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.