Honeycomb MCP vs New Relic MCP Server

Honeycomb MCP 70

The hosted server is closed, under published Terms of Service and Honeycomb Intelligence Supplemental Terms. The companion plugin is MIT (17). The privacy notice, the DPA and an AI policy page dated 21 September 2026 are public. The AI page says MCP's model provider is whichever the client uses, and that Honeycomb uses no provider that trains foundation models on input. Telemetry is deleted after the retention window, 60 days for most customers. The terms also let Honeycomb use customer data to improve its services, and the AI page says Honeycomb may fine-tune pre-trained models (22). The self-hosted server carries a deprecation notice dated 21 August 2025. No deprecation policy was found, the AI terms allow any part of Honeycomb Intelligence to be changed or ended at any time with commercially reasonable efforts at notice, and the docs say rate limits may change (6). The subprocessor list is public with purposes and countries, the DPA promises 30 days' notice of new subprocessors, and customers choose a US or EU data location (19).

New Relic MCP Server 81

Closed service with clear dated terms (paid 8 November 2022, unpaid 21 October 2021) naming New Relic, Inc. (15). A privacy notice of 14 September 2026, a data retention page with per-namespace periods, a security policy, and the MCP docs' statement that gap-analysis responses aren't stored; the master cloud agreement and the trust centre couldn't be read (22). End-of-life notices with dates (REST API keys, 2025) and a pre-release policy that says preview services can change or end (16). US, EU and JP regions are documented for the server and the platform, but no sub-processor list was found on a readable page (10). 15 + 22 + 16 + 10 = 63.

Honeycomb MCP 70

Scored as a hosted MCP server. status.honeycomb.io is an Atlassian Statuspage with components for US1, EU1, the website and the sandbox. It has no MCP component, though MCP incidents are posted there, which takes 15 as sentry-mcp and slack-mcp take for a vendor page with no MCP component (15). history.atom lists five incidents from 6 August to 4 November 2026, each limited to one region, one product or a subset of requests. US1 queries over older data erroring for 67 minutes on 10 August, Canvas not loading for 38 minutes on 21 August, degraded US event ingest for 12 minutes on 20 October, slower query execution and SLO processing in the US from 07:04 to 09:26 PDT on 20 October, and SLO, Trigger and anomaly emails failing for 15 minutes on 22 October. None took the query path down for everyone, so minor only (20). Per-tool limits are published with numbers (15). Rate-limit errors carry the time to retry after, which counts as Retry-After alone. No backoff guidance or idempotency keys for the nine write tools were found (10). No SLA found in the terms, and honeycomb.io/sla returned 404 on 9 October (0). The docs still list MCP among betas graduated to general availability (10). 15+20+15+10+0+10 = 70.

New Relic MCP Server 60

Scored as a hosted MCP server. status.newrelic.com is an Atlassian Statuspage with 115 components across the US, EU and JP regions and no MCP component (20). history.rss holds seven incidents in the 90 days to 3 November 2026, among them delayed queries and alerts in the US for 49 minutes on 2 November, synthetics ping delays the same day, the JP entity platform unavailable for 67 minutes on 28 October, delayed US data and alerts from 19:24 to 22:04 UTC on 25 September, intermittent UI errors for two and a half hours the same night, a Salesforce support outage on 16 September and a 10-minute EU metric issue. Two regional incidents over an hour on the query and alert paths the server depends on, so one major (10). Published numbers, 2,000 tool calls an hour per user on the MCP server and 25 concurrent NerdGraph requests per user (15). NerdGraph answers 429 on concurrency with guidance to batch or pool, but no Retry-After is documented and nothing covers retries of MCP calls; the tools are reads, so idempotency doesn't arise (8). The pricing page lists support response SLAs, and the paid terms promise commercially reasonable efforts, so no availability SLA (0). Public preview on 4 November 2025, available to all accounts on 3 November 2026 with preview tools marked separately, but no dated GA notice was found (7). 20 + 10 + 15 + 8 + 0 + 7 = 60.

Honeycomb MCP 68

OAuth authorisation code with PKCE S256, dynamic client registration, client ID metadata documents and refresh tokens. Since the 9 October check the consent screen lets the owner choose read all, read and write all, or a custom choice per resource, with scopes such as boards:read, triggers:write and investigations:write beside environments:read and queries:read. Connections made earlier keep mcp:read or mcp:write. The alternative is a Management API key whose scopes are fixed at creation and which can be disabled or deleted (28). Tools are read-only unless a write scope is granted, members with a read-only role can choose only read scopes, API keys are limited to chosen environments, and there is no delete tool. No confirmation step for writes was found (15). Trace, log and span content is untrusted text, and no injection guidance was found in the MCP docs (2). The Activity Log records OAuth sessions with requested and granted scopes and one event per query run with its source, MCP included. The compliance page describes activity logging as for enterprise customers (11). SOC 2 Type II each year with penetration tests by an independent firm, HIPAA BAAs and an email disclosure route under a bug bounty page whose rewards are all at Honeycomb's discretion, so the bounty isn't counted. No security.txt, no stated acknowledgement time, and the scope names ui.honeycomb.io and api.honeycomb.io, not the MCP hosts. ISO 27001 is claimed only for the cloud hosts. One certification and a disclosure route, set between mailersend's 8 (disclosure route, discretionary bounty, no certification) and daytona's 15 (disclosure route, SOC 2 and ISO 27001, discretionary reward not counted) (12).

New Relic MCP Server 76

OAuth 2.0 with the scopes observability:read, offline and offline_access, access tokens that expire after 5 minutes, refresh for 24 hours, revocation by deleting the user or turning off the organisation toggle, and the token acting as the user with no more than their roles; the alternative is a user API key in a header (28). A read scope, organisation toggles that default the API key, preview, Atlassian and AWS routes to off, an organisation-scoped role requirement, and no write tools in the list (16). Logs, error messages and trace attributes are untrusted text and no injection guidance was found (2). NrAuditEvent records sign-ins with IP addresses and NrdbQuery labels queries sent through the server, both queryable by the operator (13). security.txt pointing to a HackerOne programme, a published security policy citing SOC 2 Type 2 and ISO 27001 audits of its facilities and its own penetration testing, and an ISO 42001 certification announced on 29 September 2026; the trust centre itself renders through a script and was unread (17). 28 + 16 + 2 + 13 + 17 = 76.

Honeycomb MCP 57

The server is closed source and tools/list needs sign-in, so we could not read the input schemas. Partial credit for a contract that exists but is not public (12). docs.honeycomb.io publishes llms.txt and a Markdown twin of every page (10). The tools reference says what each of the 36 tools returns and when an agent reaches for it, for example calling get_workspace_context first and list_memories before read_memory. These are docs, not the definitions the model sees (14). Parameters are mentioned only in passing, with no types, enums or required fields published (4). A use-cases page and a troubleshooting page document the Unauthorized, rate-limit, permission and not-found errors with causes and fixes (9). The registry entry carries version 1.0.0 and Honeycomb links a UI changelog and an API changelog, but changelog.honeycomb.io still disallows unnamed crawlers, so we could not confirm MCP entries (8).

New Relic MCP Server 51

Tool input schemas are only visible through an authenticated session; the tool reference names 40 tools with one-line descriptions, so partial credit for a contract that exists but isn't public (10). docs.newrelic.com publishes llms.txt, updated 3 November 2026, and a Markdown twin of every page (10). Each tool's line says what it does, tags say which group it belongs to, and the reference says which tools need OAuth and which are preview; nothing says when not to use a tool (12). No parameters, types or enums are published; the setup examples show only the include-tags header (4). Example natural-language queries per group, a troubleshooting page for OAuth, API key, permission and connection errors, and no error code list (7). A dated what's-new feed and a 0.1.0 registry version, but no MCP changelog or server version in the docs (8). 10 + 10 + 12 + 4 + 7 + 8 = 51.

Honeycomb MCP 53

36 tools is above the 30 line (5). The X-MCP-Toolset header loads query or canvas subsets, and the docs don't say how many tools each holds (add 8, so 13). Size controls are described per tool, such as up to 100 datasets or 100 columns a call, subtree zoom on get_trace and query limits. Pagination parameters aren't documented (10). Errors are documented with causes and fixes, and the rate-limit message names the retry time. No error code list was found (12). The reference labels each tool read or write and update_slo keeps omitted fields. No idempotency keys were found, and we couldn't confirm readOnlyHint or destructiveHint annotations (8). get_workspace_context takes no parameters, relative time is the default, and setup is one command or link in nine documented clients. SDKs don't apply (10).

New Relic MCP Server 59

40 tools in the default list, above the 30-tool line (5), with the include-tags header to load only named groups (add 10, so 15). Time windows on most tools, convert_time_period_to_epoch_ms, filters on incidents, errors and logs, and NRQL's own LIMIT and FACET; no documented page size (14). The troubleshooting page covers setup errors with fixes, and the docs say nothing about tool-level error bodies (8). Every tool is a read (query, list, fetch, analyse) and the gap tool reads files the client sends, so a retry is safe; no readOnlyHint or destructiveHint was confirmed because tools/list wasn't read (12). Natural-language prompts need no parameters from the user, and the API behind the server has official NerdGraph clients; nothing agent-specific beyond skills (10). 15 + 14 + 8 + 12 + 10 = 59.