Operator's Reference · MCP Catalog · vol. I

The DataGrout
Field Manual.

A practical, opinionated index of every tool exposed over the DataGrout MCP server — written for agents that prefer brevity to ceremony, and for the humans who supervise them.

Tools indexed

116^*

Suites

Hidden by policy

Generated

2026·05·11

Concepts & primitives / prerequisites

read first · vocabulary of the rest of this manual

Tool invocation pattern across the catalog: mcp__datagrout__discovery_perform with tool_name: "data-grout@1/<suite>.<tool>@1". The vocabulary below recurs throughout.

Cache refs rc_*

Tool results stored server-side. Pass the cache_ref into the next tool's payload rather than re-sending data. Cornerstone of zero-cost chaining.

Credits

Per-call billing currency. data.* and frame.* are 0 credits (pure manipulation). LLM-backed tools - prism, invariant, warden - cost more.

CTC

Cognitive Trust Certificate. Ed25519-signed execution proof. Chains across nested calls. Produces a verifiable workflow audit trail.

Logic cell

Per-user Prolog fact store. Persistent across sessions, namespace-isolated. The reasoning substrate beneath logic.* and governor.*.

Skill

A saved, reusable workflow. Forge from a goal, invoke by id, temper to refine. Survives sessions; can be made workspace- or world-public.

Conduit SDK

Client libraries (Python, TypeScript, Rust, Elixir, Ruby). Auth: bearer, OAuth 2.1, mTLS. Transports: MCP, JSONRPC, WebSocket.

discovery / 05 tools

find · plan · execute

The bootstrap suite. Every other call begins, in spirit, with discover → perform. Use plan for multi-step traversal; summary for catalog reconnaissance.

discovery.discover

Semantic tool search across the MCP server. Returns a ranked list with input/output contracts.

goal: string · brief? · compress? · coverage_gaps? · integrations? · limit? · min_score? · servers?

discovery.guide

Guided decision-tree navigation. Multi-turn interactive discovery when the goal is fuzzy.

goal? · query? · choice? · back? · session_id?

discovery.perform

Execute a tool by name. Supports inline analyze, refract, and chart post-processing in a single call.

tool_name: string · args? · inputs? · skill_handle? · skill_id? · save_as_skill? · schedule?

discovery.plan

Plan a multi-step workflow from a goal. Returns ranked plans plus virtual-skill handles.

goal? · have? · k? · return_call_handles? · server?

discovery.summary

Compact catalog overview, focusable to a single suite. Cheapest way to keep your bearings.

focus?

invariant / 04 tools

code analysis · structural facts · goal alignment

Extracts structural and semantic facts from source code, then reasons about them. Pair with the invariant CLI for repo-scale onboarding.

invariant.code_lens

Extract structural and semantic facts from a single source file or snippet.

invariant.code_query

Query the fact-space for known smells.

invariant.diff_analyzer

Assesses whether a diff actually accomplishes its stated goal — not just whether it compiles.

goal · before? · after? · unified_diff? · constraints? · criteria? · context? · language?

invariant.review

Combined lens + diff + query in one pass. Use as a pre-merge gate.

mode?: check | gate · repo_id? · commit_sha? · constraints? · criteria? · context? · language?

Field noteFor database-backed code, supply context: { model_facts, set_construction } — otherwise the tool will hallucinate plausible-but-wrong invariants. CLI counterpart: the invariant binary (init, lens, query, diff, review, status, onboard).

warden / 04 tools

prompt injection detection · adversarial review

A three-tier defence escalating from protocol adherence to structured threat classification, with an ensemble mode that runs all three and adjudicates between them.

warden.canary

Protocol adherence check. Runs validator tasks against untrusted content — injected instructions interfere with compliance, betraying themselves.

content · effort? · expected_format? · on_fail?: warn | block | manual_review · verbose?

warden.intent

Intent and authority drift detection. The right default for user-facing inputs.

content · expected_context? · effort? · on_fail? · verbose?

warden.adjudicate

Adversarial threat classification. Answers "what kind of attack is this?" — not just whether it's suspicious.

content · intent_result? · expected_context? · effort? · on_fail? · verbose?

warden.ensemble

All three tiers, plus adjudication between them. Highest confidence; highest cost.

content · expected_context? · expected_format? · effort? · on_fail? · verbose?

data / 15 tools · zero-cost

JSON manipulation · filter · sort · map · merge

Pure transformations over JSON-shaped values. All accept a payload (literal or cache_ref) and return a fresh cache_ref — the substrate for zero-token chaining.

data.aggregate 0c

Reduce a list to a single value via a named operation.

data.count 0c

Count items in an array or keys in a map.

payload

data.drop 0c

Skip the first N items.

n: integer

data.filter 0c

Filter array items by declarative predicate.

where: object[]

data.flatten 0c

Flatten nested structure.

depth? · mode?: map | array · separator?

data.get 0c

Access value at a path in a nested structure.

path

data.keys 0c

Return keys of a map or indices of an array.

payload

data.map 0c

Call an MCP tool once per item; collect results.

tool | expr · as? · concurrency? · max_items? · on_error?: collect | skip | abort · timeout_ms?

data.merge 0c

Merge two maps; shallow or deep.

base? · target? · deep?

data.omit 0c

Drop specified keys.

keys: string[]

data.pick 0c

Keep only specified keys.

keys: string[]

data.sort 0c

Sort array by field or natural order.

by? · dir?: asc | desc

data.sum 0c

Alias for data.aggregate.

op · field?

data.take 0c

Return the first N items.

n: integer

data.unique 0c

Deduplicate; keeps first occurrence.

by? · path?

frame / 08 tools · zero-cost

table ops · group · pivot · join · slice

Treat data as a list-of-records (rows). Everything in data applies; frame adds the operations a SQL programmer expects.

frame.filter 0c

Filter rows by declarative predicates.

where: object[]

frame.group 0c

Group by one or more fields and aggregate.

by · agg?: object[]

frame.join 0c

Inner / left / right / outer join two record lists on shared key fields.

on · type?: inner | left | right | outer

frame.pivot 0c

Long → wide reshape.

column · index · value · agg?: first | last | sum | mean | list

frame.pluck 0c

Extract a single field into a flat array.

field · compact?

frame.select 0c

Select, rename, or drop columns.

columns? · drop?

frame.slice 0c

Contiguous row slice.

offset? · limit?

frame.sort 0c

Sort rows by one or more columns.

by · order?: asc | desc

math / 12 tools

statistics · sequences · trends · outliers

Deterministic numerics. The agent stops guessing at statistics and starts computing them.

math.correlate

Pearson / Spearman correlation between two numeric series.

method?: pearson | spearman | both · x · y · x_field · y_field

math.describe

Descriptive statistics: mean, median, std, percentiles, skew, kurtosis. Optional histogram.

bins? · field?

math.interpolate

Easing / interpolation ops for animation and remapping.

operation: lerp | clamp | remap | smoothstep | ease_in/out_quad/cubic/quart/sine

math.linspace

N evenly-spaced numbers between start and stop.

n: integer · endpoint? · start? · stop?

math.normalize

Scale a numeric array via zscore, minmax, or rank.

method?: zscore | minmax | rank

math.outliers

Outlier detection by IQR or z-score.

method?: iqr | zscore · k? · threshold?

math.range

Arithmetic sequence start → stop, step.

stop · start? · step?

math.rank

Assign ranks — ordinal, dense, average, or percentile.

method? · ascending? · order?

math.sample

Draw N samples from a distribution or from existing data. seed? for reproducibility.

math.sequence

Named mathematical sequence of N terms.

math.trend

Fit a regression and optionally forecast forward.

model?: linear | polynomial | exponential | logarithmic · forecast? · degree?

math.window

Sliding-window operations: moving average, cumsum, diff, pct_change, lag, EWMA.

op · window? · n? · alpha?

prism / 09 tools

natural-language transforms · charts · concept navigation

Where the catalog meets language. Refract reshapes by goal; analyze reasons over a payload; render emits prose. Orient and Horizon navigate the latent neighbourhood of any concept.

prism.refract LLM

Reshape data per a natural-language goal. Returns cache_ref.

goal · verbose?

prism.analyze LLM

Structured reasoning over a payload. Returns findings + evidence + confidence.

goal · mode?: deductive | causal | comparative | exploratory · depth?

prism.chart LLM

Render a chart from data. Returns hosted SVG URL plus a sparkline text fallback.

prism.export

Convert content to PDF, JSON, NDJSON, XML, LaTeX, HTML, Markdown, CSV, XLSX, YAML, or text.

format

prism.focus

Type bridge between incompatible semio-types.

source_type · target_type

prism.horizon

Find the boundary of a concept's latent neighbourhood — and optionally expand beyond.

seed · levels? · expand_beyond?

prism.orient

Navigate a concept axis: negation, abstraction, agency, scale, formality, narrative, custom.

axis · angle?: opposite | orthogonal | complement | 45 | custom

prism.paginate

Page through a large result.

page? · per_page? · advance?

prism.render LLM

Generate an article, report, summary, figure, or documentation from structured data.

CompositionOrient maps a neighbourhood and returns an orientation_id. Feed that id into Horizon (with expand_beyond) to reach territory the seed alone wouldn't surface.

logic / 12 tools

prolog fact store · persistent symbolic memory

A per-user Prolog cell with sub-millisecond lookup. Assert facts, write rules, query in natural language or in raw Prolog. Persists across sessions, scoped by namespace.

logic.assert no-LLM

Insert facts directly. Zero LLM cost (vs remember).

facts: object[] | prolog: string[] · dry_run? · namespace? · tag?

logic.constrain

Store a rule or policy.

rule: string · namespace? · tag?

logic.export

Dump all active facts.

namespace? · tags? · types?

logic.forget

Retract by handle, handles, pattern, or tag.

handle | handles | pattern | tag

logic.hydrate

Natural-language context hydration from memory.

goal · format?: text | structured · max_tokens? · include_constraints?

logic.import

Bulk import from a prior export.

merge? · dry_run?

logic.query

Query with NL question (3 credits), structured patterns, or raw prolog (1 credit).

question | patterns | prolog · limit? · offset? · timeout?

logic.reflect

Memory introspection.

entity? · full? · summary_only?

logic.reload

Re-compile stored rules from persistence.

namespace?

logic.remember LLM

NL fact storage.

statement | facts · mode?: append | upsert | preview · resolve? · ttl_seconds?

logic.tabulate

Facts → tabular records for frame.* consumption.

entities? · types? · columns?

logic.worlds

List all namespaces with fact and constraint counts.

—

Power userDirect Prolog: logic.query(prolog: "metric(E, \"arr\", V), V > 1000000"). Full graph-traversal predicates; 1 credit; zero LLM latency.

batteries / 05 tools

pre-built prolog rule modules

Canned rule packs — reasoning, games, business — that compose by unification in a shared namespace.

batteries.search

Search the registry.

query? · category?: reasoning | games | business · tags?

batteries.describe

Full documentation: predicates and rules.

id · namespace? · include_rules?

batteries.install_many

Install modules into a namespace.

ids: string[] · namespace

batteries.installed

List installed batteries in a namespace.

namespace

batteries.remove

Remove a battery's rules from a namespace.

id · namespace

flow / 04 tools

workflow orchestration · approval gates

Multi-step plans that the platform verifies for cycles, type compatibility, and policy compliance before execution. Each plan emits a CTC.

flow.into

Define / validate / execute a multi-step workflow.

plan: object[] · input_data? · input_schema? · validate_only? · save_as_skill? · schedule?

flow.request-approval

Human gate. Blocks until approved.

action · details · reason · context?

flow.request-feedback

Ask the user for missing fields.

missing_fields · reason · current_data? · suggestions?

flow.route

Conditional dispatch with fall-through matching.

branches: object[] · else?

scheduler / 03 tools

time-based · event-driven · sensors

Cron-style and event-driven triggers, expressed in plain language: "every weekday at 8am UTC", "whenever 10 new leads arrive in Salesforce".

scheduler.create

Schedule a tool call or workflow.

schedule · tool? · args? · plan? · deliver_to? · max_fires?

scheduler.list

List scheduled tasks.

scheduler.cancel

Cancel a scheduled task by id.

task_id

governor / 03 tools

neuro-symbolic runtime · continuous cognition

Continuous-cognition mode. Every tool call you make auto-enriches the Prolog fact space — no explicit insertion required. Backed by the paper "Governor: Neuro-Symbolic Runtime for Token-Efficient Agent Cognition".

governor.enable

Start or resume a Governor session.

—

governor.disable

End the Governor session.

—

governor.status

Dashboard view of the running session.

—

toolsmith / 04 tools

forge · temper · catalog · invoke

Mint new tools at runtime. Forge them from a goal, refine via temper, store them in the catalog, invoke them by id.

toolsmith.forge

Mint a new skill from a goal.

goal · integration? · entity? · api_type? · save_as_skill? · skill_visibility?: private | workspace | public

toolsmith.temper

Refine or fork an existing skill.

goal · mode?: update | fork · skill_id? · tool_name?

toolsmith.invoke

Execute a saved skill.

skill_id · args? · inputs?

toolsmith.catalog

Browse skills.

filter? · integration? · source_type?: forge | temper | guide | flow.into | manual · include_public?

latent / 03 tools

concept expansion · depth · breadth · bridge

Brainstorm by traversing a semantic neighbourhood. Bridge two concepts directly; expand depth-first or breadth-first.

latent.expand

Expand a seed concept into a landscape.

seed · mode?: depth | breadth | bridge · levels? · target? · output_format?

latent.horizon

Boundary of a concept's latent neighbourhood. (Same operation as prism.horizon.)

seed · levels? · expand_beyond?

latent.orient

Axis navigation. (Same operation as prism.orient.)

axis · angle?

Cost ladderLevels: 1 → ~2 credits · 2 → ~5 · 3 → ~8. Bridge mode is a flat ~5 credits regardless of level.

docs / 04 tools

persistent named documents

Working text that survives the session. Markdown, plain text, checklist, or structured JSON.

docs.create

Create a new persistent document.

title · body? · format?: markdown | text | checklist | json · tags? · metadata?

docs.get

Retrieve by ref (doc_*).

ref

docs.list

List docs with filters.

tags? · search? · format? · limit? · offset?

docs.update

Update or append.

ref · body? · append? · title? · tags? · metadata?

deliverables / 03 tools

permanent work products

Finalised output artefacts — charts, exports, reports — registered for permanent retrieval. Compare with docs.*: deliverables are done, docs are in progress.

deliverables.register

Save a work product as a deliverable.

title · type: chart | export | data | content | report · summary? · tags?

deliverables.get

Fetch by ref.

ref

deliverables.list

Search deliverables.

query? · type? · agent_id? · limit? · offset?

tasks / 05 tools

background execution control

Inspect, wait on, or cancel background tasks the platform runs on your behalf.

tasks.list

Recent background tasks.

status?

tasks.status

Current status of a task_ref.

task_ref

tasks.result

Full result of a completed task.

task_ref

tasks.wait

Block up to wait_ms for completion.

task_ref · wait_ms?

tasks.cancel

Cancel a running task.

task_ref

inspect / 03 tools

audit trail · CTC chain-of-trust

CTCs are Ed25519-signed Cognitive Trust Certificates — verifiable proof that a workflow ran as declared.

inspect.execution-history

Recent executions.

status?: success | error | timeout · refractions_only? · limit? · offset?

inspect.execution-details

Full details for an execution_id.

execution_id

inspect.ctc-executions

All executions for a CTC or saved skill.

ctc_id? · skill_id?

ephemerals / 02 tools

cache inspection · working memory

The working-memory view of DataGrout. Every cache_ref can be peeked at without re-running the tool that produced it.

ephemerals.list 0c

List all active cached datasets.

sort?: newest | expiring_soon | largest

ephemerals.inspect 0c

Stats + preview for a specific cache_ref.

cache_ref · preview_rows?

flux / 03 tools

feedback · telemetry · engagement

Tell the platform what worked. Tell it what didn't. Track your own analytics events.

flux.feedback

Provide platform feedback.

nps_score? · ease_of_use? · trust_level? · surprise_score? · liked? · improve? · would_use_voluntarily? · unique_capability? · free_text?

flux.scorecard

Rate the individual tools used in this session.

tools: object[] · feedback_ref?

flux.track

Track a custom analytics event.

event_name · event_data?

library / 02 tools

documentation lookup · reference content

The catalog's own self-documentation. Search by query; retrieve by ref.

library.search

Search DataGrout docs and research papers.

query · suite? · type?: reference | tutorial | paper | overview | all · limit?

library.get

Retrieve content by ref.

ref · goal?

lumen / 03 tools

benchmark device · LLM cost & efficiency

Records "laps" — benchmark runs annotated with token counts and provider cost in USD. Compare configurations head-to-head.

lumen.laps

List laps with cost and token counts.

limit? · sub_id?

lumen.dashboard

Cost / efficiency dashboard for a Lumen device.

days? · sub_id?

lumen.compare

Compare two benchmark laps.

lap_a · lap_b · sub_id?

· · ·

Bundled integrations / external

first-party shims to outside systems

github · 41 tools

list_pull_requests, list_releases, get_team_members, merge_pull_request, update_pull_request, delete_file, sub_issue_write, create_repository, and more.

github---datagrout-org · 41 tools

Org-scoped GitHub surface. Currently flagged stale; auth uses an access_token.

PolicyEight tools are hidden by policy: 4 PII, 2 side-effect restricted, 2 destructive. Run discovery.summary focus="disabled" to enumerate them.

Adjacent projects / not MCP tools

the ecosystem around the catalog

conduit-sdk

Client libraries in Python, TypeScript, Rust, Elixir, Ruby. Auth: bearer, OAuth 2.1, mTLS. Transports: MCP, JSONRPC, WebSocket.

invariant (CLI)

Local binary: onboard, init, lens, query, diff, review, status. Analyses Ruby, Python, TypeScript and uploads facts to DataGrout.

logic-batteries

GitHub repository for community Prolog rule modules. Installed via batteries.install_many.

tether (Roblox)

A DataGrout client for Roblox. Methods: query, assert, generate_rules, invoke, orchestrate, call, watch. Compute credits free; inference credits limited.

arbiter (internal)

Referenced in the Governor paper as "Arbiter Substrate" — an OS-level governance layer. No public repository at the canonical URL (returns 404). Treat as DataGrout-internal infrastructure.

Patterns & recipes / practitioner notes

how the catalog composes in practice

The standard recipe

# discover what's available, then run it
discovery.discover(goal) → discovery.perform(tool_name, args)

Chained data operations — zero credit

fetch → cache_ref → data.filter → data.sort → data.take → final

Reasoning over external data

fetch → prism.analyze(goal, mode: deductive) → prism.render(content_type: report)

Mint a reusable skill

toolsmith.forge(goal, save_as_skill: true) → toolsmith.invoke(skill_id)

Inline post-processing on a single call

{
  tool_name,
  args,
  analyze: { goal, mode },
  refract: "reshape result for downstream",
  chart:   "bar chart of revenue by quarter"
}

Verifiable audit chain

flow.into → produces CTC → inspect.ctc-executions(ctc_id)

The DataGroutField Manual.

Concepts & primitives / prerequisites

discovery / 05 tools

invariant / 04 tools

warden / 04 tools

data / 15 tools · zero-cost

frame / 08 tools · zero-cost

math / 12 tools

prism / 09 tools

logic / 12 tools

batteries / 05 tools

flow / 04 tools

scheduler / 03 tools

governor / 03 tools

toolsmith / 04 tools

latent / 03 tools

docs / 04 tools

deliverables / 03 tools

tasks / 05 tools

inspect / 03 tools

ephemerals / 02 tools

flux / 03 tools

library / 02 tools

lumen / 03 tools

Bundled integrations / external

github · 41 tools

github---datagrout-org · 41 tools

Adjacent projects / not MCP tools

conduit-sdk

invariant (CLI)

logic-batteries

tether (Roblox)

arbiter (internal)

Patterns & recipes / practitioner notes

The standard recipe

Chained data operations — zero credit

Reasoning over external data

Mint a reusable skill

Inline post-processing on a single call

Verifiable audit chain

The DataGrout
Field Manual.