# Fix list: rotki MCP server From Anchor Terminal's listing at https://www.anchorterminal.com/tools/rotki-mcp, the October 2026 research run, assessed 10 October 2026. Grade E, 45.6 out of 100. This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public. For a coding agent working on rotki MCP server: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published. ## 1. Security & auth, 20 out of 100, up to 14 more on the total Why it scored 20: Credential model, desktop build. None. The launcher forces ROTKI_SESSION_KEY empty outside Docker (https://github.com/rotki/rotki/blob/v1.44.1/crates/starling/src/main.rs), so the server runs without a token, and any local process that can open 127.0.0.1:4445 can call the tools while the server runs (0 of 30). The Host and Origin allowlists in server.py stop browser pages from other origins. Docker takes a bearer token that lasts at most seven days and is revoked on logout, which is a single key and is not scored here. Read-only. No tool writes, and query_sql accepts one SELECT or WITH statement with a denylist, on an in-memory SQLite connection that never writes back to rotki (15 of 20). Prompt injection. Get_asset_details returns token names and symbols from GlobalDB as they are. The code's own comment says an asset name is attacker-controlled on a scam token. No injection guidance was found in the docs (5 of 15). Per-call visibility. The MCP module logs nothing per tool call. The server log follows --log-level, and the usage event goes to rotki when the analytics setting is on (0 of 15). Security programme. No SECURITY.md in the repository, no security.txt, and no bug bounty or disclosure policy found in the sources we read. The GitHub Security tab was not confirmed (0 of 20). Total 20. The checklist (https://www.anchorterminal.com/benchmark/#checklist-security): - 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option. - 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions. - 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10. - 0 to 15, audit logs or per-call visibility for the operator. - 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public. Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing. ## 2. Payments & pricing, 10 out of 100, up to 11.3 more on the total Why it scored 10: Local software with a paid option, so the rubric scores the Premium plan. No x402, MPP or L402 (0). Plan prices are public without a login. Basic €25 a month and Advanced €45 a month, with no yearly prices shown (10 of 20, plan-only). No free tier reaches the MCP tools. Starter has no MCP access and info is the only free tool (0). A person has to take out a Premium plan and sign in to the desktop app, and no programmatic signup was found (0). Total 10. The free, self-hosted app with nothing to buy would score 20, 20 and 20 on the last three lines, but the MCP tools are paid, so that rule does not apply. The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments): The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/). - 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which. - 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login. - 20, a free tier or trial that doesn't need a card. - 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API). Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied. Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol. ## 3. Reliability, 50 out of 100, up to 10 more on the total Why it scored 50: Read with the local-software lines, as localghost was. Installs from the rotki desktop app for Windows, macOS and Linux, as README.md states (https://github.com/rotki/rotki/blob/v1.44.1/README.md), and from a Docker image built from packaging/docker (20). Public CI exists. Rotki_ci.yml runs on pushes to develop and master, and task_backend_tests.yml runs pytest over rotkehlchen/tests, which includes tests/mcp. The Actions page showed no pass or fail for the latest develop run, so passing CI is scored absent (0). Unchecked. GitHub's issue pages are disallowed for our reader, so open crash and regression issues could not be read and score absent (0). The repository page shows 389 open issues and 17 open pull requests on 10 October 2026, without dates or labels. Semver-style releases (1.43.0, 1.43.1, 1.43.2, 1.44.0 and 1.44.1) with breaking changes called out in docs/dev_changelog.rst, for example a renamed message type and a retired environment variable, although MCP tool changes are not listed there (15). Version 1.44.1, past 1.0 (15). Total 50. The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability): Hosted APIs, MCP servers, models and platforms. - 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own). - 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so. - 15, rate limits documented with numbers. - 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved. - 10, an SLA published for any paid tier. - 10, the surface agents use is generally available, not beta or preview. Local packages, SDKs, frameworks and stdio MCP servers. - 20, installs from an official package with supported runtimes stated. - 25, a public CI and test suite, passing on the default branch. - 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered). - 15, semver discipline and breaking changes called out in a changelog. - 15, version 1.0 or later, or declared stable. Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors. ## 4. Maintenance & community, 35 out of 100, up to 5.7 more on the total Why it scored 35: v1.44.1 was released on 2 October 2026, eight days before the check (30). Two releases in the 90 days to 10 October, v1.44.0 on 21 August and v1.44.1 on 2 October, against three needed (0). Responsiveness could not be read, since GitHub's issue pages are disallowed for our reader, so it scores absent (0). No rotki entry was found in the official MCP registry search, and the repository has no server.json (0). Renovate is configured (.github/renovate.json5), dependencies are pinned in uv.lock, and CI is defined, but the latest CI result was not readable (5 of 10). Total 35. The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance): - 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older. - 20, at least three releases or dated changelog entries in the last 90 days. - 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15. - 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models). - 10, package health, current dependencies and CI. Models are read for deprecation notice periods and model churn rather than release counts. ## 5. Agent ergonomics, 69 out of 100, up to 5 more on the total Why it scored 69: Context cost. Eight tools with 9,258 characters of descriptions in total, of which refresh_analytics_data is 2,959 and query_sql 2,667. There is no toolset switch or read-only subset, though the server has no write tools (20 of 25). Output sizing is by row cap, not offset. Query_sql caps results at 500 by default and 5,000 at most, and reports result_truncated and the true row_count. Refresh filters history by time, and --max-events can bound a load, which is off by default. There is no paging cursor (15 of 20). Errors carry a code and a message. Premium_required and premium_check_failed each give an upgrade URL, and SQL errors point to list_tables and describe_table (16 of 20). No readOnlyHint or destructiveHint annotations are set in rotkehlchen/mcp, and every tool is read-only by construction, so retries are safe (10 of 20). Defaults are sensible. Refresh loads history_events unless told otherwise, and few inputs are required. No official SDK for the MCP server or the REST API was found in the repository or the docs we read (8 of 15). Total 69. The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics): - 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries). - 20, pagination, filtering and output-size controls. - 20, actionable, documented error responses, codes and messages an agent can recover from. - 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations. - 15, sensible defaults, few required parameters, and official SDKs in at least two languages. Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs. ## 6. Transparency & trust, 48 out of 100, up to 4.6 more on the total Made of editorial 55, provenance 40. Why it scored 48: Source and licence. AGPL-3.0 in LICENSE.md, with the MCP module in the same repository (30). Data handling. The privacy policy at rotki.com/privacy-policy names Rotki Solutions GmbH and covers the website, log data and cookies. It does not mention the desktop app, Premium data or the MCP, and it gives no retention period. The MCP docs say data can be shared with the connected AI client. The statements do not describe what the feature sends, and the app's own strict copy does (10 of 30). Deprecation. No policy for the REST API or the MCP tools was found. The changelog announces renames with 'must migrate' and no notice period (5 of 20). Telemetry. The general settings offer 'Submit anonymous usage analytics', which defaults to on (rotkehlchen/db/settings.py). The MCP server sends a usage event with client name, client version and model to sigil.rotki.com once per session and model. The privacy policy does not describe that event, and it does not name the operator of sigil.rotki.com (10 of 20). Total 55. The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency): - 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms. - 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors). - 0 to 20, a deprecation policy or notices with dates. - 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted). The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two. Provenance checks not met in full (half of this category, computed from checked facts): - Domain age: rotki.com, no registry record we could read (0 of 15) - Endpoint on the vendor's domain: is not on rotki.com (0 of 15) - Terms of service: not found (0 of 10) - Status page: not found (0 of 10) - security.txt: could not be fetched (0 of 10) ## 7. Schema & documentation, 76 out of 100, up to 3.9 more on the total Why it scored 76: The eight tools are registered with FastMCP, and every input is a typed Python parameter, so the JSON Schema served through tools/list is typed on every tool (25). No separate OpenAPI file exists for the MCP server. llms.txt is served at rotki.com/llms.txt, but it indexes the docs and does not mention the MCP tools (10). Descriptions say when to use most tools. Refresh_analytics_data says to call it first, describe_table says to use it before writing SQL, and get_event_taxonomy says to call it before any aggregate over event_type. Use-not-when is implied rather than stated (15 of 20). Inputs are typed but carry no enums or bounds. Tables is a list of free strings with the valid names in the description, max_rows has no schema bound, and time filters accept seconds or milliseconds (6 of 15). get_event_taxonomy carries worked SQLite recipes, and errors come back as structured payloads with a message and an upgrade URL (10 of 15). Versioned releases and a public changelog in docs/changelog.rst and docs/dev_changelog.rst record API changes, but not MCP tool changes (10 of 15). Total 76. The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema): APIs and MCP servers. - 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool). - 10, llms.txt or Markdown docs served for agents. - 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference. - 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs. - 0 to 15, examples and documented error responses. - 15, versioning and a public changelog. Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference. ## What we couldn't check What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it. - unchecked: whether the latest develop CI run passed. The Actions page showed no pass or fail, so passing CI is scored absent and reliability would rise by 25 if it passed. - unchecked: open crash and regression issues, and maintainer responsiveness, because GitHub's issue pages are disallowed for our reader. The MCP issues were not read. - unchecked: any SECURITY policy or disclosure route. The GitHub Security tab was not confirmed, and rotki.com/security.txt and /.well-known/security.txt returned client errors. - unchecked: status.rotki.com, which is disallowed for our reader, and the rotki.com terms of service, which could not be loaded. - unchecked: the domain registration date for rotki.com. The official MCP registry search returned no rotki entry, so a registry listing is not established. - The founder's lead says the server minimises the data shared, and quotes an X post that says 'Share nothing'. We could not read the post. The listing does not repeat that line, and the code sends amounts, dates and counterparties in every mode, so the claim is not made in our copy and no deduction is taken. - The source was read, not run. No command was executed, and the desktop launch, the Docker route and the ChatGPT tunnel were not tested. - The code read is develop at commit 2783d11 (8 October 2026). The MCP files cited are identical at v1.44.1 except for comments and a Windows import shim (_compat.py) that is newer than the release. The launcher and privacy copy cited are from v1.44.1 and unchanged in substance. - The stdio transport exists in rotkehlchen/mcp but is not documented, and the listing records HTTP only. - The repository page rounds the star count to 4.0k, so githubStars is left null. ## Weaknesses - The desktop endpoint at http://127.0.0.1:4445/mcp takes no token, because the session key is forced empty outside Docker, so any local process can call it while the server runs - Amounts, USD values, timestamps, asset symbols, counterparties and venue names pass unmasked in every mode, as the in-app strict description says - The privacy policy at rotki.com/privacy-policy does not mention the desktop app or the MCP, and the MCP usage event to sigil.rotki.com is on by default - Two releases in the 90 days to 10 October 2026, and no rotki entry found in the official MCP registry search - The feature is seven weeks old, and no injection guidance, per-call audit log or security disclosure policy was found ## What costs an agent a turn today The notes we give agents before they call it. Each one is a workaround an agent shouldn't need. - Ask the person to start the server under Settings > MCP in the desktop app, since it stays down until started, then connect to http://127.0.0.1:4445/mcp - Call info first. It works without Premium and reports whether an unlocked backend is connected - Call get_event_taxonomy before any aggregate over event_type, and sum on direction or event_group, since informational rows double count MEV and block rewards - Call refresh_analytics_data before query_sql, and check source.completeness and unpriced_rows before reporting any total - Check the privacy mode before answering from the data. Strict keeps amounts and dates, and raw sends addresses, hashes and notes verbatim ## When it's done Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.