Danube
Governed tool access for AI agents — one Danube API key unlocks your organization's own tools plus a large, growing catalog of services, over MCP or curl, with confirmation before anything that writes, sends, spends, or deletes.
as observed 2026-10-18T21:45:32.509Z- Identifier
tools-marketplace- Source
- ClawHub
- Version observed
- 8.1.38
- Source repository
- not published
- Repository observation
- No source repository listed
- First observed here
- 2026-08-28T22:06:39.940Z
- Observations recorded
- 30
- Installs (reported upstream)
- 82
- Weekly downloads (upstream)
- 2,882
- Declared license
- MIT-0
Observation history
2026-10-18T21:45:32.509Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"A service's slug reads that service's tools but scopes a search to nothing, and the two\ncatalogue-size numbers on `/health` are equal when nothing is waiting to be embedded.\n\n- | "`retryable` is not \"every upstream failure\", and a `local_only` tool refuses instantly when the\norganization's data-plane agent is not polling.\n\n- **`SKILL.md`: the `retryabl |
latestVersion |
"8.1.37" | "8.1.38" |
2026-10-17T20:48:50.018Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"`list_services` grew a detail level that empties the rows above a limit of 25, it drops every\nservice the catalogue reports with no tools, and the `-32602` failure the bundle war | "A service's slug reads that service's tools but scopes a search to nothing, and the two\ncatalogue-size numbers on `/health` are equal when nothing is waiting to be embedded.\n\n- |
latestVersion |
"8.1.35" | "8.1.37" |
2026-10-15T18:48:12.129Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Three stale live figures re-measured, and the handshake's capabilities block is not the server's\ninventory.\n\n- **`references/troubleshooting.md` quoted FastMCP `4.0.5` as the h | "`list_services` grew a detail level that empties the rows above a limit of 25, it drops every\nservice the catalogue reports with no tools, and the `-32602` failure the bundle war |
latestVersion |
"8.1.34" | "8.1.35" |
2026-10-14T17:39:07.962Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Why a tool that fails every call can look healthy in a search result, and the failure message that\nsays a catalogue row has no code behind it.\n\n- **`usage` and `reliability` co | "Three stale live figures re-measured, and the handshake's capabilities block is not the server's\ninventory.\n\n- **`references/troubleshooting.md` quoted FastMCP `4.0.5` as the h |
latestVersion |
"8.1.33" | "8.1.34" |
2026-10-13T16:37:28.889Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Two capabilities an agent could not have found in this bundle: a workflow's own schedule, and\nthe half of `set_parameter_defaults` that decides what the save overwrites.\n\n- **W | "Why a tool that fails every call can look healthy in a search result, and the failure message that\nsays a catalogue row has no code behind it.\n\n- **`usage` and `reliability` co |
latestVersion |
"8.1.32" | "8.1.33" |
2026-10-12T15:36:43.438Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"The hosted-AI cap covers five tools, not the nine this bundle named — and the other five\n`Danube - …` names cannot be executed at all.\n\n- **The \"hosted AI\" list was wrong in | "Two capabilities an agent could not have found in this bundle: a workflow's own schedule, and\nthe half of `set_parameter_defaults` that decides what the save overwrites.\n\n- **W |
latestVersion |
"8.1.31" | "8.1.32" |
2026-10-11T14:42:02.570Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Two numbers in `/health` that are not the same number, and the one-call way to ask how big the\ncatalogue is — the second was in `SKILL.md` but never reached the `curl` reference. | "The hosted-AI cap covers five tools, not the nine this bundle named — and the other five\n`Danube - …` names cannot be executed at all.\n\n- **The \"hosted AI\" list was wrong in |
latestVersion |
"8.1.30" | "8.1.31" |
2026-10-10T14:15:52.413Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"How `execute_tool` turns a name into a tool — checked against the merged MCP server and probed\nlive, because the answer is that it sometimes runs a different tool and says so onl | "Two numbers in `/health` that are not the same number, and the one-call way to ask how big the\ncatalogue is — the second was in `SKILL.md` but never reached the `curl` reference. |
latestVersion |
"8.1.29" | "8.1.30" |
2026-10-09T13:20:37.280Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Two things a client reads as a limit and shouldn't, both checked against the merged code and\nprobed live this cycle.\n\n- **The OAuth `scope` is a label, not a permission boundar | "How `execute_tool` turns a name into a tool — checked against the merged MCP server and probed\nlive, because the answer is that it sometimes runs a different tool and says so onl |
latestVersion |
"8.1.27" | "8.1.29" |
2026-10-08T12:19:32.872Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Two things an agent reads on every search and every batch, one of them never documented and one\nof them changed under the file on 2026-10-04. Both checked against the merged code | "Two things a client reads as a limit and shouldn't, both checked against the merged code and\nprobed live this cycle.\n\n- **The OAuth `scope` is a label, not a permission boundar |
latestVersion |
"8.1.26" | "8.1.27" |
2026-10-07T11:52:53.866Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Two product changes landed on 2026-10-04 that the bundle had not caught up with, and one of them\nbreaks a rule this file states everywhere else. Everything below verified against | "Two things an agent reads on every search and every batch, one of them never documented and one\nof them changed under the file on 2026-10-04. Both checked against the merged code |
latestVersion |
"8.1.25" | "8.1.26" |
2026-10-06T10:50:34.888Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"8.1.23 printed the MCP endpoint's `initialize` reply, `serverInfo` `4.0.5` included, without\nsaying what that number is. It is the FastMCP library's version, not Danube's, and `/ | "Two product changes landed on 2026-10-04 that the bundle had not caught up with, and one of them\nbreaks a rule this file states everywhere else. Everything below verified against |
latestVersion |
"8.1.24" | "8.1.25" |
2026-10-05T10:33:23.015Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Three things an agent runs into while wiring the connection up by hand, none of them written down\nbefore: which verb the MCP endpoint answers, how to read its `401`s, and the key | "8.1.23 printed the MCP endpoint's `initialize` reply, `serverInfo` `4.0.5` included, without\nsaying what that number is. It is the FastMCP library's version, not Danube's, and `/ |
latestVersion |
"8.1.23" | "8.1.24" |
2026-10-04T09:46:17.065Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"`structuredContent` was documented as the one payload that escaped `max_response_chars`. A fix on\n2026-09-27 (`backend/execution/projection.py`, `mcp/mcp_server.py`) closed that, | "Three things an agent runs into while wiring the connection up by hand, none of them written down\nbefore: which verb the MCP endpoint answers, how to read its `401`s, and the key |
latestVersion |
"8.1.22" | "8.1.23" |
2026-10-03T08:46:10.142Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"The bundle has named the infrastructure connectors' write tools since 8.1.20, but only as far as\nthe consent handshake. What those tools refuse *after* consent was undocumented, | "`structuredContent` was documented as the one payload that escaped `max_response_chars`. A fix on\n2026-09-27 (`backend/execution/projection.py`, `mcp/mcp_server.py`) closed that, |
latestVersion |
"8.1.21" | "8.1.22" |
2026-10-02T07:46:18.975Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"The consent handshake has a second trigger, and the bundle named only the first. Everywhere it\nappeared — the guardrail, both reference files and the skill card — a `confirmation | "The bundle has named the infrastructure connectors' write tools since 8.1.20, but only as far as\nthe consent handshake. What those tools refuse *after* consent was undocumented, |
latestVersion |
"8.1.20" | "8.1.21" |
2026-10-01T06:47:50.210Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Instructions-mode workflows, which the bundle had only ever mentioned in passing: how to create\none, where its answer actually is, the three ways its orchestrator fails, and why | "The consent handshake has a second trigger, and the bundle named only the first. Everywhere it\nappeared — the guardrail, both reference files and the skill card — a `confirmation |
latestVersion |
"8.1.19" | "8.1.20" |
2026-09-30T05:46:36.081Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Workspaces and sharing with an organization, plus three side routes that now treat a tool you\ncannot see the way they treat one that isn't there, and a JSON body Danube used to m | "Instructions-mode workflows, which the bundle had only ever mentioned in passing: how to create\none, where its answer actually is, the three ways its orchestrator fails, and why |
latestVersion |
"8.1.18" | "8.1.19" |
2026-09-29T04:47:03.526Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Context replaces the user profile, and skills can belong to an organization.\n\n- **`identity://user` is gone, and so is `GET /v1/identity/api`.** In their place is context:\n ma | "Workspaces and sharing with an organization, plus three side routes that now treat a tool you\ncannot see the way they treat one that isn't there, and a JSON body Danube used to m |
latestVersion |
"8.1.17" | "8.1.18" |
2026-09-28T03:46:57.521Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"An MCP tool listing is a summary now, and three refusals an agent would otherwise misread.\n\n- **`search_tools` and `get_service_tools` over MCP no longer carry the full schema.* | "Context replaces the user profile, and skills can belong to an organization.\n\n- **`identity://user` is gone, and so is `GET /v1/identity/api`.** In their place is context:\n ma |
latestVersion |
"8.1.16" | "8.1.17" |
2026-09-27T02:45:50.156Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Three behaviour changes an agent reads differently, from Danube #882, #883 and #884.\n\n- **A structured result is sent once.** `execute_tool` and `fetch_result` used to repeat th | "An MCP tool listing is a summary now, and three refusals an agent would otherwise misread.\n\n- **`search_tools` and `get_service_tools` over MCP no longer carry the full schema.* |
latestVersion |
"8.1.14" | "8.1.16" |
2026-09-26T01:46:07.395Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Every failed tool call now tells an agent whose failure it was, and the skill had no idea: this release documents the `fault` / `error_type` / `retryable` fields, the five reliabi | "Three behaviour changes an agent reads differently, from Danube #882, #883 and #884.\n\n- **A structured result is sent once.** `execute_tool` and `fetch_result` used to repeat th |
latestVersion |
"8.1.12" | "8.1.14" |
2026-09-25T00:46:42.104Z
Fields that differed: license changelog description latestVersion
| Field | Before | After |
|---|---|---|
license |
null | "MIT-0" |
changelog |
"Catches the skill up with a week of search and execution changes an agent can see: search results now say whether a tool is deprecated, name it by slug, and pin an exact name firs | "Every failed tool call now tells an agent whose failure it was, and the skill had no idea: this release documents the `fault` / `error_type` / `retryable` fields, the five reliabi |
description |
"Governed tool access for your agent — one Danube API key unlocks your organization's own tools plus a large, growing catalog of ready-made services. Search, inspect, and execute t | null |
latestVersion |
"8.1.11" | "8.1.12" |
2026-09-10T12:18:05.764Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Documents three things the platform started telling agents that the skill didn't pass on — a projection that matched nothing, the credential fields a service actually wants, and w | "Catches the skill up with a week of search and execution changes an agent can see: search results now say whether a tool is deprecated, name it by slug, and pin an exact name firs |
latestVersion |
"8.1.10" | "8.1.11" |
2026-09-05T08:21:05.203Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Corrects a claim about Danube's own AI tools that stopped being true the day after it was published, adds the two ways those tools can now refuse a call, and drops a warning about | "Documents three things the platform started telling agents that the skill didn't pass on — a projection that matched nothing, the credential fields a service actually wants, and w |
latestVersion |
"8.1.9" | "8.1.10" |
2026-09-04T08:09:22.338Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Documents two platform behaviours that shipped after 8.1.5 was written, both of which change what an agent sees coming back from `execute_tool` — and one of which an agent followi | "Corrects a claim about Danube's own AI tools that stopped being true the day after it was published, adds the two ways those tools can now refuse a call, and drops a warning about |
latestVersion |
"8.1.6" | "8.1.9" |
2026-09-03T07:20:35.503Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"Documents four live behaviours the skill had never mentioned, each of which an agent following 8.1.4 would have misread — three of them as a failure that wasn't one.\n\n- **The fr | "Documents two platform behaviours that shipped after 8.1.5 was written, both of which change what an agent sees coming back from `execute_tool` — and one of which an agent followi |
latestVersion |
"8.1.5" | "8.1.6" |
2026-09-02T06:49:02.991Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"- Major documentation overhaul: an all-new SKILL.md thoroughly explains Danube’s governed tool access, guardrails, API key setup, and usage workflows.\n- Clarifies security practi | "Documents four live behaviours the skill had never mentioned, each of which an agent following 8.1.4 would have misread — three of them as a failure that wasn't one.\n\n- **The fr |
latestVersion |
"8.1.2" | "8.1.5" |
2026-08-31T05:56:02.339Z
Fields that differed: changelog latestVersion
| Field | Before | After |
|---|---|---|
changelog |
"- Listing presentation only — no change to the skill instructions. Added `agents/openai.yaml` (display name \"Danube\", short description, icon) so ClawHub shows the proper name a | "- Major documentation overhaul: an all-new SKILL.md thoroughly explains Danube’s governed tool access, guardrails, API key setup, and usage workflows.\n- Clarifies security practi |
latestVersion |
"8.1.1" | "8.1.2" |
Correction
If you maintain this extension and believe anything above is inaccurate, request a correction. Corrections are published, and disputed entries are marked as disputed while under review.