recon

Passive domain intelligence from public sources. recon reads public DNS,
certificate transparency, and unauthenticated Microsoft and Google identity
discovery endpoints to compose typed observations around a domain's public
technology and identity namespace. A domain is the query coordinate, not proof
of one organization, owner, account, or deployed product.

It uses no credentials, no API keys, no paid feeds, and no active scanning. It
ships as a local Python package with a CLI, versioned JSON output, and a stdio
MCP server.
It is not a hosted service, scheduler, vulnerability scanner, company research
tool, or firmographic database.

Defensive use only. Use recon for legitimate posture review, IT
architecture review, vendor diligence, and defensive hardening. See
docs/legal.md for the intended-use policy.

Quick Start

Install with uv or pipx:

uv tool install recon-tool
# or
pipx install recon-tool

Python 3.11 through 3.14 is supported. The latest Python 3.14 patch is
recommended for new installations and development; older supported versions
retain the same product behavior and output contracts. Current measurements
and version-specific decisions are in
docs/performance.md.

Optional helpers at scripts/install.ps1 and scripts/install.sh use an
existing uv or pipx installation. Download a
release-tag source archive,
review the local helper, and then run it. Each helper installs the exact version
represented by that tag, preserves a sole existing uv or pipx owner, and
refuses ambiguous or unmanaged installations. Review the helper from a newer
tag before using it to update. Do not pipe mutable branch content directly into
a shell.

For a provenance-first path that verifies the exact GitHub and PyPI artifacts,
their signed evidence, and their byte parity before optional installation, use
the cross-platform consumer verification recipe.
The release workflow in this source revision signs the completed SBOM with the
distributions and validates any existing release's exact identity and
expected-only inventory before replacing assets during recovery. The published
v2.6.3 bundle predates the SBOM subject expansion; the quick path binds that
single historical exception to its exact commit and still validates the SBOM.

Open a new terminal and run an offline verification of the installed command:

recon --version

Optionally test online connectivity to recon's public data sources:

recon doctor

The default diagnostic uses synthetic requests to Microsoft identity endpoints,
DNS for example.com, and crt.sh. It does not query a user-supplied target.
The --fix, --mcp, and --client modes are local-only.

Before the first lookup, note that recon makes DNS queries which recursive and
authoritative DNS infrastructure may observe. Its only default request to a
target-owned HTTP endpoint is the standards-defined MTA-STS policy fetch at
https://mta-sts.<domain>/.well-known/mta-sts.txt. Google CSE and BIMI direct
probes run only when --direct-probes is explicitly enabled.

Run the first lookup:

recon example.com

If a lookup times out or every online source fails, run recon doctor to
check source connectivity, then retry the lookup.

Example output shape:

Synthetic Example Namespace
example.com

Provider     Microsoft 365 (MX delivery path) + Proofpoint gateway (MX delivery path)
Tenant       a1b2c3d4-e5f6-7890-abcd-ef1234567890
Auth         Federated
Confidence   High (4 sources)

Services
  Email       Microsoft 365, Proofpoint, DMARC, DKIM, SPF strict
  Identity    Okta, Entra ID
  Cloud       Cloudflare, AWS Route 53

Insights
  Federated identity observed; identity-vendor indicators: Okta
  Email security: observed controls: DMARC reject, DKIM, SPF strict, BIMI
  MX gateway observed: Proofpoint

Examples use IETF reserved namespaces. Tenant IDs, services, and domains in
this rendered example are fabricated. No real organization is depicted as the
evaluated target.

For detailed install, update, uninstall, and first-run workflows, read
docs/getting-started.md.

What recon Is Good For

recon reports observations, not verdicts. A missing DMARC record is a missing
record. A Microsoft 365 tenant indicator is an observed indicator. The operator
decides what those facts mean in context.

Common Commands

recon example.com                              # default panel
recon https://www.example.com/path             # normalize URL to apex
recon mail.example.com                         # reduce sub-host to apex
recon mail.example.com --exact                 # keep that literal host
recon example.com --explain                    # retained evidence and explanation
recon example.com --full                       # expanded evidence, domains, posture
recon example.com --plain                      # linear text for screen readers and grep
recon example.com --json                       # structured lookup record
recon batch domains.txt --json                 # batch JSON array
recon batch domains.txt --ndjson               # one record per line
recon batch domains.txt --summary              # aggregate-only cohort summary
recon delta example.com                        # diff against cached snapshot
recon cache show                               # bounded payload-free cache overview
recon cache show --all                         # inspect every completed JSON cache file
recon cache show mail.example.com --exact      # inspect literal-host result and CT cache metadata
recon cache clear mail.example.com --exact     # clear literal-host CT and result cache keys
recon mcp install --client=cursor              # wire MCP into a client
recon doctor --mcp                             # static MCP registry check
recon mcp doctor                               # live MCP tools/resources check
recon doctor --client=cursor                   # verify the client config

The default cache overview enumerates filenames but opens at most the
lexicographically first 100 result files and 100 CT files. It reports exact
file and uninspected counts; --all opts into complete JSON payload
inspection. Cache-writer-shaped interrupted-write artifacts are counted
without reading their payload, make the overview exit 4, and are removed by the confirmed
recon cache clear --all workflow.

These list, search, and show catalog commands are local and make no network
requests:

recon fingerprints list                       # summarize public-record matchers
recon fingerprints search email               # preview matching fingerprint slugs
recon fingerprints show <slug>                 # show every record and rule for one slug
recon signals list                             # list derived reportable observations
recon signals search email                     # search signal definitions
recon signals show "<signal name>"             # inspect one signal definition

A fingerprint slug is a named public-record indicator. One slug can
intentionally have multiple catalog records, and show preserves every record,
detection, relationship hint, tier, and verification date. A signal is a
derived, reportable observation definition built from candidate fingerprint
slugs and metadata conditions. Human fingerprint search previews at most ten
unique slugs; --json returns every matching catalog record. Category filters
use the same word-prefix or phrase rule in the CLI and MCP server. Human catalog
detail output strips terminal controls from locally extended text and visibly
marks values truncated after 1,024 characters.

At terminal widths below 70 columns, command help automatically switches to a
complete linear layout so long option names and command summaries remain
visible without source markup. The no-argument welcome also switches to
vertical command descriptions whenever an aligned row would not fit, including
at ordinary widths. --plain is the linear view for a standard single-domain
lookup; batch, delta, chain, compare, exposure, and gaps reports keep their
mode-specific formats.

Built-in posture profiles: fintech, healthcare, saas-b2b,
high-value-target, public-sector, and higher-ed. Custom profiles live in
~/.recon/profiles/*.yaml.

Generated command and flag reference:
docs/cli-surface.md.

How recon Works

recon reads:

• DNS records: MX, TXT, SPF, DMARC, DKIM, BIMI, CNAME, NS, SRV, and CAA.
• Certificate transparency: SAN names, issuers, issuance timing, and bounded
  related-domain hints.
• Identity discovery: unauthenticated Microsoft and Google endpoints.

Default collection includes bounded DNS queries through the configured
recursive resolver, so authoritative DNS infrastructure may observe resulting
resolver traffic. The only default target-owned HTTP or application request is
the standards-based MTA-STS policy fetch at mta-sts.<domain>. Google CSE and
BIMI VMC direct probes are opt-in behind --direct-probes.

The engine then maps observables to fingerprint slugs, derived signals, typed
topology, provenance paths, per-slug evidence strength, and model-relative
Bayesian diagnostics. Sparse public evidence stays sparse: the result lowers
confidence or remains unresolved instead of inventing a clean answer. A source
failure remains unavailable rather than becoming a negative observation. The
Bayesian uncertainty band is evidence-responsive, not a demonstrated credible
or confidence interval.

The reviewed built-in fingerprint source remains split YAML. Release wheels
load one deterministic generated JSON catalog, while custom and session-scoped
fingerprints still pass through the runtime validator. This keeps contributor
review readable and removes repeated YAML parsing from cold CLI startup without
changing catalog order, matching, or public output.

Long-form explanation:
docs/how-it-works.md.
Formal model and robustness research program:
docs/correlation.md.

JSON and Automation

recon <domain> --json emits a stable single-domain lookup object. Batch and
delta modes emit different shapes, so route by mode or by record_type.
Retained batch modes create at most the requested number of outer batch worker
tasks while restoring input order. Each admitted lookup may still create its
own bounded source tasks. --ndjson also releases completed records and remains
the lowest-memory choice for large inputs.
Batch processing is record-oriented: a valid run keeps exit code 0 when one or
more domains fail. JSON and NDJSON consumers must inspect record_type and
error_kind on every record; CSV consumers must inspect the error column.
recon batch --summary --json preserves the separate aggregate-only
cohort_summary 2.1 contract. New consumers can select
--summary-schema 2.2 for raw-evidence-bound DMARC rates, corrected missingness,
and explicit metric kinds. The standalone reducer uses
--schema-version 2.2 for the corresponding atemporal compatibility view.

recon example.com --json
recon batch domains.txt --json
recon batch domains.txt --ndjson
recon delta example.com --json

Read these before building an integration:

• docs/schema.md: stable JSON contract.
• docs/recon-schema.json: machine-readable schema.
• docs/automation-examples.md: parser examples.
• docs/operational-contract.md: timeouts,
  bounds, exit codes, cache, and partial-result semantics.

docs/surface-inventory.json, docs/cli-surface.md, and
recon://surface-inventory are generated discovery context and drift guards,
not stable runtime API contracts. ADR-0007 records the promotion gate for any
future stable subset.

MCP Server

The default install includes a local stdio MCP server for MCP-compatible tools.
Start with manual approvals. Approval syntax is client-specific, and some
current client schemas do not define autoApprove. Treat connected agents as
untrusted input.

recon mcp install --client=claude-desktop
recon doctor --mcp
recon mcp doctor
recon doctor --client=claude-desktop

The three checks cover different boundaries in order: static server registry,
live local stdio discovery plus canonical tool and JSON resource reads, then
the named client's saved configuration. The live check does not inspect client
configuration. The installer writes the right per-client config shape and
preserves sibling MCP servers. Full setup, tool list, read-only versus stateful
guidance, and troubleshooting live in
docs/mcp.md.
Per-client scaffolds live in
agents/.

Limitations

The public channel has a ceiling:

• Internal-only workloads are invisible.
• SaaS products without DNS verification records may not appear.
• Email gateways can hide the downstream mailbox provider.
• CT logs can be stale, partial, rate-limited, or absent.
• Fingerprints are rule-based indicators, not proof of active use.

Read docs/limitations.md
before using recon output for a high-stakes decision. Read
docs/data-handling-policy.md
before committing any validation artifact.

Documentation

• docs/getting-started.md: install, update,
  uninstall, and first commands.
• docs/how-it-works.md: readable model overview.
• docs/README.md: complete docs index.
• docs/roadmap.md: current plan, invariants, and scope
  boundaries.
• docs/structural-maintainability.md: measured
  source, test, compatibility, and facade cleanup plan.
• docs/external-writeup-plan.md: active
  maintainer plan for external write-up readiness.
• docs/submission-freeze-checklist.md:
  final paper and artifact freeze gate before any external submission package.
• docs/c3-ct-validation-plan.md: closed
  certificate-transparency validation plan.
• CHANGELOG.md: shipped changes.

Roadmap Focus

recon has a stable baseline, but product quality work remains. The top three
priorities are:

1. Make every default claim traceable to evidence and remove product-use,
   cloud-type, or security-maturity conclusions that public metadata cannot
   support.
2. Keep the exact MCP v1.28.1 and v2.0.0b1 compatibility matrix green, then
   repeat the full gate against the final 2026-07-28 specification and stable
   v2 SDK before changing the production dependency.
3. Establish an aggregate-safe quality baseline for claim precision,
   abstention, provenance, catalog coverage, degradation, latency, CT value,
   and agent context cost before expanding inference or graph machinery.

Catalog coverage work uses deduplicated private rounds across rank, region, and
domain-class strata. Real target names and per-domain records stay in ignored
local validation workspaces. GitHub receives only generic provider patterns,
provider-owned references, reserved synthetic fixtures, and disclosure-safe aggregates.
The opt-in maintenance path now accounts separately for CNAME chains, apex
CNAME, non-SPF TXT, SPF targets, MX, NS, CAA, DMARC RUA, bounded owner-qualified
TXT, and bounded SRV observations. A frozen aggregate baseline and distinct
sampling rounds are required before the project claims broad catalog coverage.
The frozen convenience-sample baseline and a 366-namespace unseen vertical
holdout are complete and public as aggregate counts only; independent rank,
regional, vendor-seed, and drift rounds remain open.

The dependency order, acceptance evidence, stop rules, and current code-graph
summary live in
docs/roadmap.md.
The concise execution sequence is the
Quality Proof plan.
The implementation plan is
docs/engineering-refinement-plan.md, and
the current step-back review is
docs/strategic-gap-audit.md.
Research publication,
OpenSSF process, outside replication, and archive work remain separate
maintainer tracks so they do not displace product truthfulness or measured
utility.

The most recent completed historical local proof for the separate publication
track is
validation/2026-06-30-submission-freeze-local-proof.md.
The current paper and artifact package is unfrozen after subsequent product,
documentation, and release changes. Maintainers must rerun the submission gate
before external submission.
Its
public-label decision
keeps public
lists as robustness checks rather than population rates, and its
M365 tenancy decision
keeps that evidence as
corroboration rather than independent calibration.

Development

uv sync
uv run pre-commit install
uv run python scripts/release_readiness.py --allow-dirty
uv run python scripts/check.py

uv run python scripts/check.py is the canonical local gate. It runs lint,
type checks, coverage-gated tests, fingerprint and generated-artifact checks,
validation and added-line text hygiene, tracked Markdown link and heading-anchor
validation, workflow and dependency-export guards, interface checks, paper
claim and figure checks, and size/complexity ratchets. Its full-suite stage uses
at most four file-grouped test workers while preserving combined branch
coverage. Focused pytest commands stay serial by default. Do not push on
--fast alone. CI and the release workflow pass an explicit Git revision range
so text hygiene covers every added line in the pushed or release commit range,
not only the final diff.

Release-shaped package builds use uv 0.11.17 and the exact hashed backend graph
in build-constraints.txt. The build creates the sdist first and constructs the
wheel from that file; the replay commands and remaining environment limits are
documented in docs/supply-chain.md.

Project hygiene: keep examples reserved and synthetic, keep validation artifacts
aggregate-only, and run uv run python scripts/check.py. The gate rejects the
retired target-example vocabulary across public text while preserving provider
definitions and ACME protocol terms; avoid dead code or placeholders.
Contributor details:
CONTRIBUTING.md.

License

Apache 2.0. Free to use, build on, fork, and share. See
LICENSE for
the full terms.