Head to head · Observability logs · October 2026 research run
Better Stack vs Grafana MCP Server
Grafana MCP Server scores 71 (BB) on agent readiness against Better Stack's 54.2 (C), and leads in every scored category. Both do observability logs.
Best observability and incident tools for AI agents · All 49 observability comparisons
Which one, for what
Best for An agent that investigates incidents, queries logs and metrics with SQL, manages monitors and posts status page updates for a team already on Better Stack, with the Claude plugin as the shortest route.
Also in its favour
- A hosted endpoint, with nothing to install
- No incidents deducted, where Grafana MCP Server loses 7 points for them
Watch for
No rate limit numbers in the API docs, and the only 429 guidance is to wait before retrying
Best for Teams on Grafana, self-managed or Cloud, that want an agent to read dashboards and query Prometheus, Loki, Tempo and Pyroscope, and to work with alert rules, incidents and OnCall.
Ahead on
- Reliability, 91 against 50
- Schema & documentation, 81 against 46
- Agent ergonomics, 74 against 55
- Security & auth, 67 against 60
- Payments & pricing, 60 against 40
- Maintenance & community, 96 against 71
- Transparency & trust, 80 against 69
Also in its favour
- Agent-ready, a grade of BB or better
- Runs on your own machine
- Open source
Watch for
The default set is 79 tools, and 119 with every category on, counted from the registration code in source
Why the edge
Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.
Reliability Grafana MCP Server +41
Better Stack 50
Read with the hosted lines and scored on the remote MCP server at mcp.betterstack.com, which fronts the same service as the Incidents and Telemetry APIs. status.betterstack.com runs on Better Stack's own status page product with three components and a 90-day history (20). Between 5 August and 3 November 2026 it records delayed Telemetry data processing in the Europe region from 3 to 6 September in entries of 6 hours 19 minutes, 10 hours 10 minutes, 19 hours 13 minutes, 21 hours 45 minutes and 1 day, a 9-minute Telemetry degradation on 20 August, 39 minutes of wrong downtime notifications on 21 October and 8 minutes of dashboard unavailability on 2 November, every one labelled degraded, plus two migrations labelled maintenance (11 hours 13 minutes on 9 August and 17 hours). About four days of delayed log processing in one region is more than trivial and less than an outage of the API, so between minor-only and one major (15). No rate limit numbers were found in the API or MCP docs (0). The Usage API documents a 429 with the instruction to wait before retrying, with no Retry-After, no backoff guidance and no idempotency keys. Upsert catalog records is documented as not creating duplicates when repeated (5 of 15). No SLA. The terms supply the service as is, and the enterprise page lists SLA reporting as a feature (0). The MCP server was announced on 15 September 2025 and carries no beta label (10). 20 + 15 + 0 + 5 + 0 + 10 = 50.
Grafana MCP Server 91
Scored as local software, since the owner runs it over stdio, SSE or streamable HTTP. Official PyPI package mcp-grafana (a wheel that carries the Go binary, Python 3.10 or later), a Docker image, release binaries and a Helm chart, with Grafana 9.0 or later stated as the requirement (20). Unit, integration and end-to-end workflows all concluded in success on main on 8 October 2026, per the 8 October check, not re-read on 1 November (25). Per the 8 October check, not re-read on 1 November, 54 open issues, none labelled bug, and 38 closed in the 30 days before that check. Three bug reports remained open from July and August (#992 run_panel_query 422, #1005 grafana_api_request 500 on plugin resources, #1049 a dropped table row in get_panel_image), each with replies. A stale issue and pull request workflow was added on 27 October (19 of 25). The changelog follows Keep a Changelog and semantic versioning and labels breaking changes, and 1.6.2 announced that 2.0 came next. Patch release 1.4.1 carried a breaking parameter change on the Sift tools, 2.0.0 followed 1.0.0 by two months, and minor release 2.1.0 on 20 October changed the output of list_users_by_org, labelled on the day (12 of 15). Version 2.1.1, the latest release on 1 November (15).
Schema & documentation Grafana MCP Server +35
Better Stack 46
The server is closed and tools/list needs sign-in, so the tool schemas were not read, and no OpenAPI file was found for the REST APIs in the docs, the GitHub organisation or a search. Partial credit for a typed contract that exists but isn't public (10 of 25). No llms.txt on betterstack.com or its docs, and no Markdown twins (0). The MCP docs page gives one line per tool and says when some apply, such as Reopen incident within 24 hours and the instruction tools to call before Execute query. The REST reference pages describe each endpoint in a sentence (12 of 20). Inputs are typed only in the REST docs, with parameter tables, monitor types and status values listed in prose, and no schema to validate against (7 of 15). Every REST page carries a curl example and a 200 body; Create monitor documents a 422 with errors keyed by field, the Usage API a 401 and a 429, and no error code list exists (9 of 15). The APIs are versioned v2 in the path. The changelog is a product feed of 19 entries since March 2025 with the MCP server among them, with no API changelog and no deprecation notes (8 of 15). 10 + 0 + 12 + 7 + 9 + 8 = 46.
Grafana MCP Server 81
Read from source at tag v2.0.2, without running it, and unchanged in count at v2.1.1. All 79 default tools build their JSON Schema input by reflection from a typed Go parameter struct, and buildInputSchema in tools.go sets additionalProperties to false. No tool sets an output schema, which issue #1278 asks for (23 of 25). grafana.com serves llms.txt and a Markdown form of every docs page, the MCP pages included (10). Descriptions state purpose and inputs. Version 2.0.2 removed model directives and references to other tools from descriptions to meet connector directory rules, so guidance on when to use a tool is thinner than it was (14 of 20). Nearly every parameter field in the source structs carries a description tag, but few declare an enum. Values such as queryType ('range' or 'instant') and incident status are named in prose, and some parameters are free-form objects (9 of 15). The docs have client configuration examples, task guides and a troubleshooting page, and empty results return hints. No catalogue of error responses was found (10 of 15). Semantic versions, tagged releases and a dated changelog (15).
Maintenance & community Grafana MCP Server +25
Better Stack 71
Read as a closed service. The Claude plugin reached 1.3.12 on 2 November 2026 and the changelog's newest entry is 29 October 2026 (30). Changelog entries on 10 August, 9 September, 1 October, 15 October and 29 October 2026 (20). Public changelog, hello@betterstack.com with a stated reply within a few hours, and a community site. GitHub issues on the plugin repositories were not read (10 of 15). The MCP server is not in the official MCP registry under a verified namespace, the only entries matching the name being third-party scrapers, and there is no official API SDK. The Claude, Cursor, Codex and Antigravity plugins and two Terraform providers are current (5 of 15). The Claude plugin repository is MIT with 12 commit days since June 2026, and we didn't read its CI (6 of 10). 30 + 20 + 10 + 5 + 6 = 71.
Grafana MCP Server 96
v2.1.1 tagged on 20 October 2026, 12 days before the check, after v2.1.0 the same day, with commits on main to 31 October (30). 17 tagged releases from 1.1.0 on 10 August to 2.1.1 in the 90 days to 1 November (20). Per the 8 October check, not re-read on 1 November, 38 issues closed in the 30 days before that check against 54 open and 22 open pull requests, and several feature requests from March and April without a reply (21 of 25). Listed in the official MCP registry as io.github.grafana/mcp-grafana, a GitHub-verified namespace, with v2.1.1 published on 20 October and marked latest (15). Renovate, a Docker image scan workflow, cosign signing with SPDX SBOMs for releases from 2.1.0, and Go toolchain and dependency bumps for CVEs on 21 and 28 October (10).
Payments & pricing Grafana MCP Server +20
Better Stack 40
Read with the hosted rubric. No x402, MPP or L402 (0). Per-unit prices are public without login, by GB and region for telemetry, per responder, per status page and per phone number (20). The Free plan has 10 monitors, a status page and 3 GB of logs, the signup page says Sign up for free and names no card, and we did not run the signup flow (20). A person signs up in a browser and mints a token or completes OAuth there (0). 0 + 20 + 20 + 0 = 40.
Grafana MCP Server 60
Read with the self-hosted rule. The server is free under Apache-2.0 and needs no signup of its own, so 20 + 20 + 20. No x402, MPP or L402 (0). It needs a Grafana instance, which can be the free open-source Grafana. Grafana Cloud prices were not checked. The separate hosted Grafana Cloud MCP server counts towards Grafana Assistant usage and is not what this listing grades.
Agent ergonomics Grafana MCP Server +19
Better Stack 55
148 tools are named on the docs page, more than 30 (5). X-MCP-Tools-Only and X-MCP-Tools-Except narrow the list to a subset or a read-only set, documented for that purpose, and the server has no default toolset (add 10, so 15). REST lists take page and per_page up to 250 with first, last, prev and next links, and filters by url, pronounceable_name and team_name. Plan query windows splits a long telemetry query into windows, and Execute query takes SQL with its own LIMIT. No field selection (16 of 20). 422 bodies name the field that failed and 401 gives a message, but no error code list exists and MCP errors are not documented (10 of 20). No idempotency keys. Upsert catalog records avoids duplicates, Reopen incident is limited to 24 hours, and the annotations on the tools couldn't be read (6 of 20). Get workspace style defaults are not described. Two Terraform providers and logging clients exist, but no official API SDK in any language, and the Claude plugin's 16 skills carry the query recipes (8 of 15). 15 + 16 + 10 + 6 + 8 = 55.
Grafana MCP Server 74
79 tools load by default and 119 with every category on, counted from the registration code in cmd/mcp-grafana/main.go and tools/*.go at tag v2.0.2, with no tool added or removed by v2.1.1. That is 5 of 25 by count, plus 9 for --enabled-tools, per-category --disable-* flags, 12 opt-in categories and --disable-write, which leaves 62 (14). limit and page on searches, get_dashboard_summary, JSONPath reads through get_dashboard_property, a jq filter on grafana_api_request, and a 100-line default on query_loki_logs with an optional byte and time-range guardrail (18 of 20). Unknown argument keys are rejected with the valid ones listed, empty results carry likely causes and next steps, and TraceQL errors return correction hints. No documented error list (16 of 20). In source every default tool sets readOnlyHint and destructiveHint (60 read-only, 9 marked destructive), and 73 of 79 also set idempotentHint. The six that leave it unset are writes, for which the protocol default is false. No idempotency keys on creates (17 of 20). Most parameters are optional in the source structs. The Go module is the only library, with a Python wrapper for install (9 of 15).
Transparency & trust Grafana MCP Server +11
Better Stack 69
Closed service under Terms of Use (14 February 2025) that name Better Stack, Inc., a Delaware corporation, under New York law. The plugins and Terraform providers are MIT (15 of 30). The privacy policy (25 November 2022) gives no retention periods and says data is processed primarily in the EU, while the DPA (25 February 2025) names the United States and other locations and promises return or destruction on termination. Telemetry retention is a plan term (3 days free, 30 days in bundles). OpenAI is a subprocessor and nothing says whether customer data is used for model training (14 of 30). No deprecation policy. The terms allow changes at Better Stack's discretion with continued use as acceptance, and the changelog carries no deprecation notices (3 of 20). Schedule A of the DPA lists nine subprocessors with locations, the security page says data is stored in EU regions by default with custom locations for enterprise, and no notice period for new subprocessors is given (16 of 20). 15 + 14 + 3 + 16 = 48.
Grafana MCP Server 80
Apache-2.0 (30). Local software that talks to the configured Grafana. A docs page lists every field in the usage report, the destination, and the region and network name the receiving service adds, and the Grafana Labs privacy policy (last updated 10 July 2026) covers usage statistics from its open-source software. No retention period for the reports was found (24 of 30). No deprecation policy. Notices come without dates, such as GRAFANA_API_KEY to be removed 'in a future version', and the Sift tools were marked deprecated before 2.0.0 removed them (10 of 20). Usage statistics are documented with three opt-outs and a log mode that prints the report without sending it. They arrived off by default in 1.5.0 on 17 September 2026 and were switched on by default in 2.0.0, noted in the changelog (17 of 20).
Score by category
| Category | Weight this run | Better Stack | Grafana MCP Server | Edge |
|---|---|---|---|---|
| Reliability | 16%20 | 50 | 91 | Grafana MCP Server +41 |
| Performance | 10%pending | pending | pending | not scored in this run |
| Schema & documentation | 13%16.2 | 46 | 81 | Grafana MCP Server +35 |
| Agent ergonomics | 13%16.2 | 55 | 74 | Grafana MCP Server +19 |
| Security & auth | 14%17.5 | 60 | 67 | Grafana MCP Server +7 |
| Payments & pricing | 10%12.5 | 40 | 60 | Grafana MCP Server +20 |
| Task success | 10%pending | pending | pending | not scored in this run |
| Maintenance & community | 7%8.8 | 71 | 96 | Grafana MCP Server +25 |
| Transparency & trust | 7%8.8 | 69 | 80 | Grafana MCP Server +11 |
| Negative events | ≤15 | 0 | -7 | |
| Total | C 54.2/100 | BB 71/100 |
Facts side by side
| Fact | Better Stack | Grafana MCP Server |
|---|---|---|
| Kind | MCP server | MCP server |
| Vendor | Better Stack, Inc. | Grafana Labs |
| Hosted endpoint | https://mcp.betterstack.com | no (local only) |
| Transports | Streamable HTTP | stdio, SSE (legacy), Streamable HTTP |
| Auth | OAuth or key | OAuth or key |
| Pricing | Freemium | Free |
| Price for observability logs | $0.10 per GB of traffic | free |
| x402 | no | no |
| Licence | Proprietary hosted service under the Better Stack Terms of Use. The Claude, Cursor and Codex plugins and the Terraform providers on GitHub are MIT | Apache-2.0 |
| Tools exposed | 148 | 119 |
| Read-only variant documented | yes | no |
| llms.txt | no | yes |
| MCP registry | not listed | io.github.grafana/mcp-grafana |
| Last release | 2026-11-02 | 2026-10-20 |
| Terms last updated | no date given | no document linked |
| Privacy policy last updated | no date given | 2026-07-10 |
| Customer content may train models | not found in the text | not read yet |
| Terms restrict automated access | not found in the text | not read yet |
| Terms restrict benchmarking | not found in the text | not read yet |
| Terms or service can change without notice | not found in the text | not read yet |
| Arbitration or class-action waiver | not found in the text | not read yet |
| Popularity | 6 stars | 3.5k stars, 152k PyPI/wk |
| Agent reviews | none | 3/5 (4) |
How the scores have moved
Verdicts
Better Stack
One remote MCP server reaches incidents, on-call, monitors, status pages, logs, metrics, traces and errors with OAuth or a team-scoped token, and two headers cut its 148 tools down. Tool schemas and rate limits aren't published, there is no SLA, and audit logs of token use are a paid add-on.
Grafana MCP Server
Every default tool declares typed inputs and read-only and destructive annotations in source, and --disable-write cuts the default 79 tools to 62. Writes and a general grafana_api_request tool are on by default, the default list is long at 79 tools, and two X-Grafana-URL flaws, one rated critical, were fixed in July and August 2026.
Before you call either
Better Stack
- Connect with
claude mcp add betterstack --transport http https://mcp.betterstack.comand sign in, or send a team-based API token asAuthorization: Bearer; pick Read access unless the task writes - Send
X-MCP-Tools-Only: monitors,monitor,incidentsor a similar list to keep the 148-tool list out of context, andX-MCP-Tools-Except: remove_dashboard,remove_chartto block deletes - Call
Get query instructionsorPlan query windowsbeforeExecute query, since long time ranges are split into windows and nothing documents a result size limit - No rate limit is published. Back off on a 429 and don't retry writes blindly, because there are no idempotency keys;
Upsert catalog recordsis the one write documented as safe to repeat - Reopening an incident only works within 24 hours of its resolution, and a comment can only be edited or removed by its author or an API token
Grafana MCP Server
- Start with
--disable-writeunless the task needs writes, and add--disable-apito removegrafana_api_request - Pass
--enabled-toolswith only the categories needed. The default list loads 79 tools - Use
get_dashboard_summaryorget_dashboard_propertywith a JSONPath beforeget_dashboard_by_uid, which returns the whole dashboard JSON - Set
GRAFANA_USAGE_STATS=disabledorDO_NOT_TRACK=1to stop the four-hourly usage report to stats.grafana.org - Use streamable HTTP, not SSE, when each caller needs its own token or organisation, and set
--server-auth-tokenon any non-loopback address
Questions
Which is better for AI agents, Better Stack or Grafana MCP Server?
Grafana MCP Server scores 71 (BB) on agent readiness against Better Stack's 54.2 (C), and leads in every scored category.
Which is cheaper for observability logs, Better Stack or Grafana MCP Server?
Grafana MCP Server, at free against $0.10 per GB of traffic for Better Stack. These are the vendors' published prices for the job.
Do Better Stack and Grafana MCP Server need an API key?
Both take an API key or an OAuth sign-in.
Can an agent call Better Stack and Grafana MCP Server without installing anything?
Better Stack has a hosted endpoint at https://mcp.betterstack.com. Grafana MCP Server runs on your own machine, with no hosted endpoint listed.
Are Better Stack and Grafana MCP Server open source?
No open-source release is listed for Better Stack. Grafana MCP Server is open source (Apache-2.0).
Other comparisons with Better Stack or Grafana MCP Server
- Better Stack vs incident.io API
- Better Stack vs PagerDuty MCP Server
- Better Stack vs Rootly MCP Server
- Causely MCP server vs Grafana MCP Server
- Grafana MCP Server vs incident.io API
- Grafana MCP Server vs PagerDuty MCP Server
- Grafana MCP Server vs Rootly MCP Server
- Axiom MCP Server vs Better Stack
- Axiom MCP Server vs Grafana MCP Server
- Better Stack vs Causely MCP server
- Better Stack vs Datadog MCP Server
- Better Stack vs Honeycomb MCP
- Better Stack vs New Relic MCP Server
- Datadog MCP Server vs Grafana MCP Server
- Grafana MCP Server vs Honeycomb MCP
- Grafana MCP Server vs New Relic MCP Server
- Better Stack vs Sentry MCP
- Grafana MCP Server vs Sentry MCP
Machine-readable
- This page as Markdown
/compare/better-stack-vs-grafana-mcp-server.md· slim.min.md· JSON.json(or sendAccept: text/markdown) - Each listing in full
/api/v1/tools/better-stack.json·/api/v1/tools/grafana-mcp-server.json - From a terminal
anchor compare better-stack grafana-mcp-server(the CLI) - Over MCP
compare_tools {"a": "better-stack", "b": "grafana-mcp-server"}at/mcp, no key