Skip to main content
Glama
eSentire Atlas MCP Server

eSentire Atlas MCP Server

by omichelbraga

eSentire Atlas MCP Server

An MCP server exposing the eSentire Atlas API (Findings, Ticketing, MVS) and Threat Intelligence feeds (IP Watch, Advanced/STIX) as tools.

Built for HTTP transport so it can run as a container behind a reverse proxy and be consumed by remote MCP clients.

Credentials

Atlas issues four distinct credential types and they are not interchangeable — presenting the wrong type returns 402/403:

Atlas credential type

Env var

Covers

Atlas

ATLAS_API_TOKEN

/finding, /tickets, /mvs — read and write

Atlas Readonly

ATLAS_API_TOKEN

same paths, reads only

Threat Intelligence

ESENTIRE_TI_TOKEN

/ti/ipwatch, /ti/indicators

GenAI

—

not implemented (undocumented endpoints)

Create them in Atlas → Settings → New API Credentials. Choose Authentication: Token — the static-token flow is the one the Atlas API Reference Guide documents. Optionally add an IP restriction for your Docker host's egress address.

Related MCP server: cti-mcp-server

Configuration

Variable

Default

Purpose

ATLAS_API_TOKEN

—

Atlas credential (Findings / Ticketing / MVS)

ESENTIRE_TI_TOKEN

—

Threat Intelligence credential

ESENTIRE_API_BASE

https://api.esentire.com

API root

ESENTIRE_CUSTOMER_CODE

—

Default tenant code, so callers can omit it

ESENTIRE_ENABLED_SURFACES

findings,tickets,ti,mvs

Comma list; drop mvs if not licensed

ESENTIRE_READ_ONLY

false

1 refuses every mutating call at the door

MCP_TRANSPORT

http

http or stdio

MCP_HOST / MCP_PORT

0.0.0.0 / 3000

Bind address

MCP_PUBLIC_URL

—

External URL; required for OAuth

MCP_AUTH

none

none | bearer | oauth/azure | github

MCP_BEARER_TOKEN

—

Shared secret when MCP_AUTH=bearer

AZURE_TENANT_ID / AZURE_CLIENT_ID / AZURE_CLIENT_SECRET

—

Entra app for MCP_AUTH=oauth

TI_CACHE_TTL

3600

Seconds. Matches the feed's hourly refresh

MAX_RESPONSE_CHARS

40000

Per-tool response ceiling

LOG_LEVEL

info

MCP_AUTH is the front door — how MCP clients authenticate to this server. It is unrelated to the eSentire tokens, which authenticate this server upstream.

Tools

Findings — findings_search, findings_search_advanced, finding_get, finding_update

findings_search returns every state unless you pass states. Atlas on its own returns only Open findings when the query has no STATE predicate, which makes a week of resolved detections look like an empty week. finding_get and finding_update take either the finding UUID or the CS case number.

Ticketing — tickets_case_type_configs, tickets_list_cases, tickets_get_case, tickets_create_case, tickets_update_case, tickets_list_comments, tickets_list_emails, tickets_list_attachments, tickets_get_attachment_link, tickets_upload_attachment, tickets_delete_attachment, tickets_list_contacts, tickets_list_locations

Threat Intelligence — ti_ipwatch, ti_check_ip, ti_indicators, ti_indicators_paged, ti_indicators_misp

MVS — mvs_list_assets, mvs_get_asset, mvs_asset_details, mvs_get_asset_vulnerability, mvs_list_vulnerabilities, mvs_get_vulnerability, mvs_list_missing_patches, mvs_assets_affected_by

Design notes

  • Filters are native objects. Atlas wants URL-encoded JSON for query, sorts, and filters. Pass real lists/dicts; encoding is handled internally.

  • assignee_email implies use_v2. Atlas silently ignores the assignee filter unless the V2 query engine is active, so findings_search enables it for you rather than returning quietly-wrong results.

  • STIX is flattened by default. ti_indicators reduces bundles to a compact indicator/observable list. Pass summarize=False for raw STIX 2.1.

  • MVS auto-switches GET → POST when the encoded filter payload gets long enough to risk a query-string limit.

  • MVS is separately licensed. Its tools are registered by default, but they return a "wrong token type / not entitled" message if the service is not on your account. Drop mvs from ESENTIRE_ENABLED_SURFACES to hide them.

  • page and per_page are mandatory on /finding/findings. Undocumented in the Atlas guide, but Atlas returns 400 Missing required query parameter without them, so findings_search always sends both (defaults 1 / 50).

  • Findings and MVS use different envelopes. Findings returns {count, items, total_count}; MVS returns {data, paging}.

  • 401 bodies are surfaced verbatim. "<CUSTOMERCODE> is not authorized" means the token is valid but the service is not subscribed; a bare "Unauthorized" means the token itself was rejected. Very different fixes.

  • Trailing slashes are stripped. Atlas returns 404 for a URL ending in /.

  • Tokens are sent raw in Authorization — no Bearer prefix.

  • /health never calls upstream. An expired token or an eSentire outage must not make Docker restart-loop an otherwise-healthy container.

Local development

uv venv --python 3.12
uv pip install -e .
MCP_TRANSPORT=stdio ATLAS_API_TOKEN=... uv run python -m esentire_mcp

Deployment

Deployed as a Portainer stack that builds this repo directly on the Docker host — see docker-compose.yml. No local Docker or image registry required.

The service joins two networks: its own esentire_net, and an external mcp-edge that the reverse proxy also sits on, so the proxy can resolve esentire-atlas-mcp:3000 by name. Create it once with docker network create mcp-edge if it does not exist.

Behind a reverse proxy, streamable HTTP needs buffering disabled and long timeouts, or long-lived responses get cut:

proxy_buffering off;
proxy_read_timeout 3600s;
proxy_send_timeout 3600s;
chunked_transfer_encoding on;

Set MCP_PUBLIC_URL to the external HTTPS URL. It is required for OAuth and otherwise only used for advertising the server's own address.

Safety

tickets_create_case opens a real support case with eSentire's SOC. Confirm with a human before calling it. Use case_type: "Atlas API Test" for smoke tests. Closed tickets can never be reopened; resolved tickets can — prefer resolving. Set ESENTIRE_READ_ONLY=1 to disable all mutations.

Resolve a detection with finding_update (state: "Resolved" plus a resolution_strategy); that resolves the linked case as well. Resolving a case directly with tickets_update_case needs state: "resolved", a comments text and a username; Atlas rejects the request otherwise. Both tools re-read the record afterwards and report whether the change actually took.

Available Tools

29 tools
findings_search_advancedA

POST variant of the Findings search, for filter payloads too large for a query string.

Inputs: body (required). Returns: result

finding_updateC

Update an existing Finding (WRITE).

Inputs: updates (required), finding_id (required). Returns: result

mvs_asset_detailsB

List a sub-resource of one asset.

Inputs: asset_id (required), resource (required), limit, sorts, +3 more. Returns: result

mvs_assets_affected_byA

Find every asset affected by a given CVE, vulnerability, or missing patch.

This is the blast-radius question: given one finding, what is exposed? Supply exactly one of cve_id, vulnerability_id, or the patch_name/patch_type pair.

Inputs: limit, sorts, cve_id, offset, +5 more. Returns: result

mvs_get_assetA

Get full detail for one asset.

Inputs: asset_id (required). Returns: result

mvs_get_asset_vulnerabilityA

Get one vulnerability as it applies to one specific asset.

Inputs: asset_id (required), vulnerability_id (required). Returns: result

mvs_get_vulnerabilityA

Get details for one vulnerability, by internal id or by CVE.

Inputs: cve_id, vulnerability_id. Returns: result

mvs_list_assetsB

List discovered assets and their vulnerability counts.

Asset records use UPPERCASE field names (ASSET_ID, HOSTNAME, THREAT_RATING, VULN_CRITICAL_COUNT, ...). Note there are two parallel count families: SOFTWARE_VULN_*COUNT and VULN*_COUNT.

Inputs: limit, sorts, offset, filters, +2 more. Returns: result

mvs_list_missing_patchesA

List missing patches across the whole estate.

Inputs: limit, sorts, offset, filters, +1 more. Returns: result

mvs_list_vulnerabilitiesA

List the vulnerability catalogue across the whole estate.

Inputs: limit, sorts, offset, filters, +2 more. Returns: result

ti_check_ipA

Check whether a specific IP address is on the eSentire IP Watch blocklist.

Inputs: ip (required). Returns: result

tickets_case_type_configsA

List the valid case_type / case_subtype / service combinations.

Always call this before creating a case. The taxonomy is server-driven and changes without notice, so hardcoding subtypes leads to rejected creates.

Inputs: case_type. Returns: result

tickets_create_caseA

Create a new case in eSentire's ticketing system (WRITE -- reaches the real SOC).

This opens a genuine support case with eSentire. Confirm intent with a human before calling it. Validate case_type/case_subtype/service against tickets_case_type_configs first; invalid combinations are rejected.

Inputs: case_type (required), short_description (required), contact, service, +11 more. Returns: result

tickets_delete_attachmentA

Delete an attachment from a case (WRITE -- not reversible).

Inputs: file_uid (required), ticket_id (required). Returns: result

tickets_get_caseA

Get one case in full.

Inputs: ticket_id (required), fields. Returns: result

tickets_list_attachmentsB

List the file attachments on a case.

Inputs: ticket_id (required). Returns: result

tickets_list_casesB

List cases from eSentire's ticketing system.

Inputs: limit, sorts, offset, states, +4 more. Returns: result

tickets_list_commentsA

List all comments on a case.

The case object itself only carries the latest 5 comments; use this for the rest.

Inputs: ticket_id (required), limit, fields, offset. Returns: result

tickets_list_contactsA

List authorised contacts for your account.

Use this to get an exact contact value for tickets_create_case, since an unresolvable contact is silently dropped rather than rejected.

Inputs: limit, offset. Returns: result

tickets_list_emailsB

List the email correspondence attached to a case.

Inputs: ticket_id (required), limit, fields, offset. Returns: result

tickets_list_locationsA

List locations registered for your account.

Same rationale as contacts: an unresolvable location is silently dropped.

Inputs: limit, offset. Returns: result

tickets_update_caseA

Update a case: add comments, set customer fields, or resolve it (WRITE).

Resolving is reversible; closing is not. Prefer resolve.

Inputs: updates (required), ticket_id (required). Returns: result

tickets_upload_attachmentA

Attach a file to a case (WRITE).

Only handles the inline path, i.e. files <= 5MB. For anything larger, Atlas requires a two-step presigned upload that has to be driven from a real filesystem, not through an MCP tool call.

Inputs: file_name (required), ticket_id (required), content_base64 (required), username. Returns: result

ti_indicatorsA

Fetch the most recent Indicator Bundles from the Threat Intelligence feed.

Covers IPs, domains, URLs, file hashes and email addresses curated by eSentire's Threat Response Unit. Upstream refreshes hourly, and results are cached for that long, so calling this repeatedly inside an hour returns identical data.

Inputs: summarize. Returns: result

ti_indicators_mispB

Fetch a MISP-format indicator file from the Threat Intelligence feed.

Inputs: year (required), file_name (required). Returns: result

ti_indicators_pagedA

Page through Indicator Bundles from the Threat Intelligence feed.

Use this instead of ti_indicators when you need to walk the whole feed without blowing the response ceiling.

Inputs: limit, offset, summarize. Returns: result

ti_ipwatchA

Fetch the eSentire IP Watch blocklist.

Returns a newline-delimited list of malicious IP addresses, suitable for pushing into a firewall or WAF blocklist.

Inputs: contains, max_entries. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 29 tools

    Disambiguation4/5

    The major domains are cleanly separated by tickets_, ti_, findings/finding_, and mvs_ prefixes, and most tools target a distinct resource+action. The only mild ambiguity comes from near-variant tools such as findings_search vs findings_search_advanced and ti_indicators vs ti_indicators_paged vs ti_indicators_misp, but their descriptions explain the differences.

    Naming Consistency4/5

    Most names follow a verb_noun pattern under a clear domain prefix, e.g. tickets_list_cases, mvs_get_asset, and ti_check_ip. A few deviations such as tickets_case_type_configs, mvs_asset_details, mvs_assets_affected_by, and finding_update alongside findings_search prevent a perfect score, but the style is otherwise consistent and readable.

    Tool Count3/5

    29 tools feels heavy for an MCP surface and will consume meaningful agent context, placing the server above the ideal 3-15 range. The count is partly justified by the broad eSentire platform scope spanning cases, threat intel, findings, and MVS, but it is still borderline and could reasonably be split into separate servers.

    Completeness4/5

    The surface covers the core case lifecycle, findings search/update, threat-intel feeds, and MVS asset/vulnerability/patch workflows well. Minor gaps include no direct single-finding fetch and no support for attachments over 5MB, but agents can usually work around these using the existing search and metadata tools.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      B
      quality
      C
      maintenance
      Enables MCP clients to interact with SentinelOne's cybersecurity platform for security analysis, threat investigation, and asset management through natural language queries. Provides read-only access to alerts, vulnerabilities, misconfigurations, and inventory data.
      33
      104
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      Provides threat intelligence tools like IoC lookups, event backtracking, and IP enrichment via MCP, enabling automated triage and evidence queries.
      1
      -
    • A
      license
      A
      quality
      C
      maintenance
      Exposes Tenable security operations as MCP tools for AI-powered security workflows, enabling asset, vulnerability, scan, plugin, and tag management via natural language.
      17
      MIT