feat: add MCP server for docs, policies, and source code#207
feat: add MCP server for docs, policies, and source code#207ac12644 wants to merge 4 commits intoNVIDIA:mainfrom
Conversation
wscurran
added
Getting Started
|
Thanks for working on the MCP server, @ac12644 — having docs, policies, and source code accessible that way is a cool idea. Quick note: main has evolved quite a bit since this PR was opened (CI pipeline, new features, etc.). We'd love to consider this, but a rebase onto the latest main would really help us give it a thorough review. Mind updating when you have a moment? |
ac12644
force-pushed
the
feat/mcp-docs-server
branch
from
f93021c to
4234a3d
Compare
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
✅ Files skipped from review due to trivial changes (1)
🚧 Files skipped from review as they are similar to previous changes (1)
📝 WalkthroughWalkthroughA new MCP server "nemoclaw-docs" is added; it indexes repository docs, policy presets, source/config files and exposes 12 MCP tools for listing, reading, searching, and retrieving blueprint/policy/models and architecture/Dockerfile content over stdio JSON-RPC. Changes
Sequence DiagramsequenceDiagram
participant Client as MCP Client
participant Server as MCP Server\n(nemoclaw-docs)
participant FS as Repository\nFileSystem
Client->>Server: initialize (MCP handshake)
Server->>FS: scan docs/, policies/, sources, root files
FS-->>Server: file contents
Server->>Server: build in-memory indexes (docs, presets, sources, models)
Server-->>Client: initialized
Client->>Server: tools/list
Server-->>Client: list of 12 tools
Client->>Server: call tool (e.g., search | read_doc | get_nim_models)
Server->>Server: query indexes or read fallback files (with path-deny checks)
Server-->>Client: structured result or isError response
Estimated Code Review Effort🎯 4 (Complex) | ⏱️ ~50 minutes Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
🧹 Nitpick comments (4)
mcp-docs-server/index.js (1)
242-253: Consider adding minimum length validation for search query.The zod schema allows empty strings for
query, which combined with the search function issue could cause problems.💡 Suggested fix
{ - query: z.string().describe("Search query (keywords or phrase)"), + query: z.string().min(1).describe("Search query (keywords or phrase)"),🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@mcp-docs-server/index.js` around lines 242 - 253, The search tool's Zod schema currently allows empty queries because it uses z.string() for the "query" field; update the schema used in server.tool for the "search" tool to enforce a minimum length (e.g., use z.string().nonempty() or z.string().min(3)) so empty strings are rejected and validation fails early before calling the search handler; modify the schema where the "query" property is defined to include .nonempty() or .min(n) to prevent empty queries.test/mcp-docs-server.test.js (1)
202-214: Version test doesn't verify server-reported version.The test reads
package.jsondirectly but doesn't assert that the server'sinitializeresponse contains the matching version. The comment on line 208 acknowledges this limitation.💡 Suggestion to verify server version
it("server version matches package.json", () => { - const res = mcpCall("initialize", { - protocolVersion: "2024-11-05", - capabilities: {}, - clientInfo: { name: "test", version: "1.0" }, - }); - // The init response is id 1, but we sent it as id 2 in mcpCall wrapper - // Use a direct check instead const pkg = JSON.parse( fs.readFileSync(path.join(ROOT, "mcp-docs-server", "package.json"), "utf-8") ); - assert.equal(pkg.version, "0.1.0", "package.json version mismatch"); + // Get the initialize response (id 1) from a fresh server call + const res = mcpCall("tools/list"); // triggers initialize with id=1 + // Parse output to find the initialize response + // For now, just verify package.json is consistent + assert.equal(pkg.version, "0.1.0", "package.json version mismatch"); });🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@test/mcp-docs-server.test.js` around lines 202 - 214, The test currently only reads package.json (pkg) but never checks the server's initialize response; update the test to capture the return value from mcpCall("initialize", ...) and assert that the server-reported version equals pkg.version. Concretely, call mcpCall("initialize", ...) into a variable (e.g., initRes), extract the server-reported version from the response payload returned by mcpCall (the field under the initialize result, e.g., initRes.result.serverInfo.version or initRes.result.version depending on the response shape), and replace the lone assert.equal(pkg.version, "0.1.0") with assert.equal(serverVersion, pkg.version) to verify they match.mcp-docs-server/README.md (1)
48-60: Consider removing leading$from console commands.The markdownlint MD014 rule flags dollar signs before commands when no output is shown. This is a stylistic preference to improve copy-paste UX.
📝 Suggested fix
```console -$ cd /path/to/NemoClaw -$ node mcp-docs-server/index.js +cd /path/to/NemoClaw +node mcp-docs-server/index.jsThe server communicates over stdin/stdout using the MCP JSON-RPC protocol.
Install Dependencies
-$ cd mcp-docs-server -$ npm install +cd mcp-docs-server +npm install</details> <details> <summary>🤖 Prompt for AI Agents</summary>Verify each finding against the current code and only fix it if needed.
In
@mcp-docs-server/README.mdaround lines 48 - 60, Remove the leading "$ "
prompt from the shell command lines in the README code blocks so they are plain,
copy-pastable commands; specifically update the command strings "cd
/path/to/NemoClaw", "node mcp-docs-server/index.js", "cd mcp-docs-server", and
"npm install" to have no leading dollar sign or space.</details> </blockquote></details> <details> <summary>mcp-docs-server/package.json (1)</summary><blockquote> `1-23`: **Consider adding `"private": true` to prevent accidental npm publish.** This internal package is unlikely to be published to npm. Adding `"private": true` prevents accidental publishes. The specified dependency versions (`@modelcontextprotocol/sdk` ^1.12.1 and zod ^3.24.0) are valid and recent. Note that zod v4.3.6 is available; if your code is compatible, consider updating to ^4.x for the latest features and improvements. <details> <summary>🤖 Prompt for AI Agents</summary>Verify each finding against the current code and only fix it if needed.
In
@mcp-docs-server/package.jsonaround lines 1 - 23, Add "private": true to the
package.json to prevent accidental npm publish; update the top-level object (the
same place as "name", "version", "description") and set "private": true so npm
will refuse publishing this internal package.</details> </blockquote></details> </blockquote></details> <details> <summary>🤖 Prompt for all review comments with AI agents</summary>Verify each finding against the current code and only fix it if needed.
Inline comments:
In@mcp-docs-server/index.js:
- Around line 439-444: The mapping that builds rows from catalog.models uses
m.minGpuMemoryMB without guarding against null/undefined, causing NaN or runtime
errors; update the rows generation (the map that produces the string forrows)
to defensively read m.minGpuMemoryMB (and optionally m.name/m.image) and fall
back to a safe default (e.g., 0 MB or "N/A") before dividing/formatting so
(Number(m.minGpuMemoryMB) || 0)(or an "N/A" branch) is used and the resulting
string is always valid.- Around line 481-493: The fallback read uses path.join(ROOT, file_path) and
readIfExists without verifying the resolved path, allowing path traversal;
update the code around path.join/ROOT/file_path/readIfExists to resolve the
target path (use path.resolve or path.relative) and ensure the resolved path is
inside ROOT (e.g., reject if path.relative(ROOT, resolvedPath) starts with '..'
or if resolvedPath does not start with ROOT), then only call readIfExists on
validated paths and return the existing "File not found" error response for any
disallowed/outside paths; reference the symbols path.join, ROOT, file_path,
readIfExists, and sourceFiles to locate and change this logic.- Around line 161-176: The search function can infinite-loop when query yields
an empty term (terms = [""]) because content.indexOf("") returns 0 repeatedly;
update the logic in search to guard against empty terms by filtering out empty
strings from terms (e.g., terms =
query.toLowerCase().split(/\s+/).filter(Boolean)) or early-return an empty array
when query.trim() is empty, and/or skip the indexOf loop when term === "" (check
term before the while using content.indexOf); ensure the change references the
existing symbols: search, terms, content.indexOf, idx so the fix is applied
where the score and while loop are computed.- Line 14: The current ROOT computation uses new URL(import.meta.url).pathname
which can produce a leading slash on Windows; replace that with Node's
fileURLToPath(import.meta.url) to get a proper platform-correct file path, then
pass that into path.dirname and path.resolve to compute ROOT (update the
expression that defines ROOT to use fileURLToPath(import.meta.url) instead of
new URL(import.meta.url).pathname and ensure fileURLToPath is imported from
'url').
Nitpick comments:
In@mcp-docs-server/index.js:
- Around line 242-253: The search tool's Zod schema currently allows empty
queries because it uses z.string() for the "query" field; update the schema used
in server.tool for the "search" tool to enforce a minimum length (e.g., use
z.string().nonempty() or z.string().min(3)) so empty strings are rejected and
validation fails early before calling the search handler; modify the schema
where the "query" property is defined to include .nonempty() or .min(n) to
prevent empty queries.In
@mcp-docs-server/package.json:
- Around line 1-23: Add "private": true to the package.json to prevent
accidental npm publish; update the top-level object (the same place as "name",
"version", "description") and set "private": true so npm will refuse publishing
this internal package.In
@mcp-docs-server/README.md:
- Around line 48-60: Remove the leading "$ " prompt from the shell command lines
in the README code blocks so they are plain, copy-pastable commands;
specifically update the command strings "cd /path/to/NemoClaw", "node
mcp-docs-server/index.js", "cd mcp-docs-server", and "npm install" to have no
leading dollar sign or space.In
@test/mcp-docs-server.test.js:
- Around line 202-214: The test currently only reads package.json (pkg) but
never checks the server's initialize response; update the test to capture the
return value from mcpCall("initialize", ...) and assert that the server-reported
version equals pkg.version. Concretely, call mcpCall("initialize", ...) into a
variable (e.g., initRes), extract the server-reported version from the response
payload returned by mcpCall (the field under the initialize result, e.g.,
initRes.result.serverInfo.version or initRes.result.version depending on the
response shape), and replace the lone assert.equal(pkg.version, "0.1.0") with
assert.equal(serverVersion, pkg.version) to verify they match.</details> --- <details> <summary>ℹ️ Review info</summary> <details> <summary>⚙️ Run configuration</summary> **Configuration used**: Path: .coderabbit.yaml **Review profile**: CHILL **Plan**: Pro **Run ID**: `e6257fe7-efc9-4ee7-aefc-d7a888b53635` </details> <details> <summary>📥 Commits</summary> Reviewing files that changed from the base of the PR and between b1fb0133b0b3ab6faf38f9f78f304a5fd7b2da7c and 4234a3d7f03f82005aa6d9777e6236f00dcdb8bc. </details> <details> <summary>⛔ Files ignored due to path filters (1)</summary> * `mcp-docs-server/package-lock.json` is excluded by `!**/package-lock.json` </details> <details> <summary>📒 Files selected for processing (5)</summary> * `.gitignore` * `mcp-docs-server/README.md` * `mcp-docs-server/index.js` * `mcp-docs-server/package.json` * `test/mcp-docs-server.test.js` </details> </details> <!-- This is an auto-generated comment by CodeRabbit for review status -->
ac12644
force-pushed
the
feat/mcp-docs-server
branch
from
4234a3d to
fd496d1
Compare
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@mcp-docs-server/index.js`:
- Around line 560-564: The nimRows generation maps nimCatalog.models and assumes
m.minGpuMemoryMB is present, which can produce NaN; update the mapping in the
nimRows creation to defensively handle missing or non-finite m.minGpuMemoryMB
(e.g., check Number.isFinite(m.minGpuMemoryMB) or m.minGpuMemoryMB != null) and
use a safe fallback (like 0 or "N/A") before calling toFixed; reference the
nimRows variable and the m.minGpuMemoryMB property so the change is applied to
the map callback that formats `m.name` and GPU memory.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 00b1ddcd-1a34-4198-8f60-899f36b017b3
⛔ Files ignored due to path filters (1)
mcp-docs-server/package-lock.jsonis excluded by!**/package-lock.json
📒 Files selected for processing (5)
.gitignoremcp-docs-server/README.mdmcp-docs-server/index.jsmcp-docs-server/package.jsontest/mcp-docs-server.test.js
✅ Files skipped from review due to trivial changes (3)
- .gitignore
- mcp-docs-server/package.json
- test/mcp-docs-server.test.js
| const nimRows = nimCatalog?.models | ||
| ? nimCatalog.models | ||
| .map((m) => `| \`${m.name}\` | ${(m.minGpuMemoryMB / 1024).toFixed(0)} GB |`) | ||
| .join("\n") | ||
| : "| (catalog unavailable) | |"; |
There was a problem hiding this comment.
Missing null check for minGpuMemoryMB in architecture overview.
The get_nim_models tool (line 445) defensively handles missing minGpuMemoryMB, but this code path does not, risking NaN output.
🛡️ Proposed fix for consistency
const nimRows = nimCatalog?.models
? nimCatalog.models
- .map((m) => `| \`${m.name}\` | ${(m.minGpuMemoryMB / 1024).toFixed(0)} GB |`)
+ .map((m) => `| \`${m.name || "unknown"}\` | ${m.minGpuMemoryMB ? (m.minGpuMemoryMB / 1024).toFixed(0) : "?"} GB |`)
.join("\n")
: "| (catalog unavailable) | |";📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| const nimRows = nimCatalog?.models | |
| ? nimCatalog.models | |
| .map((m) => `| \`${m.name}\` | ${(m.minGpuMemoryMB / 1024).toFixed(0)} GB |`) | |
| .join("\n") | |
| : "| (catalog unavailable) | |"; | |
| const nimRows = nimCatalog?.models | |
| ? nimCatalog.models | |
| .map((m) => `| \`${m.name || "unknown"}\` | ${m.minGpuMemoryMB ? (m.minGpuMemoryMB / 1024).toFixed(0) : "?"} GB |`) | |
| .join("\n") | |
| : "| (catalog unavailable) | |"; |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@mcp-docs-server/index.js` around lines 560 - 564, The nimRows generation maps
nimCatalog.models and assumes m.minGpuMemoryMB is present, which can produce
NaN; update the mapping in the nimRows creation to defensively handle missing or
non-finite m.minGpuMemoryMB (e.g., check Number.isFinite(m.minGpuMemoryMB) or
m.minGpuMemoryMB != null) and use a safe fallback (like 0 or "N/A") before
calling toFixed; reference the nimRows variable and the m.minGpuMemoryMB
property so the change is applied to the map callback that formats `m.name` and
GPU memory.
|
Thanks @cv ! Just rebased onto the latest main, should be ready for review now. |
CI fix needed: missing
|
ac12644
force-pushed
the
feat/mcp-docs-server
branch
from
fd496d1 to
2b09053
Compare
Add an MCP (Model Context Protocol) server that exposes NemoClaw documentation, blueprint config, network policies, NIM model catalog, and source code as 12 searchable tools for AI assistants. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Abhishek Chauhan <ac12644@gmail.com>
ac12644
force-pushed
the
feat/mcp-docs-server
branch
from
2b09053 to
c661b07
Compare
There was a problem hiding this comment.
♻️ Duplicate comments (1)
mcp-docs-server/index.js (1)
560-564:⚠️ Potential issue | 🟡 MinorMissing null check for
minGpuMemoryMBin architecture overview.The
get_nim_modelstool (line 445) defensively handles missingminGpuMemoryMB, but this code path does not, riskingNaNoutput. This inconsistency was flagged in a prior review.🛡️ Proposed fix for consistency
const nimRows = nimCatalog?.models ? nimCatalog.models - .map((m) => `| \`${m.name}\` | ${(m.minGpuMemoryMB / 1024).toFixed(0)} GB |`) + .map((m) => `| \`${m.name || "unknown"}\` | ${m.minGpuMemoryMB ? (m.minGpuMemoryMB / 1024).toFixed(0) : "?"} GB |`) .join("\n") : "| (catalog unavailable) | |";🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@mcp-docs-server/index.js` around lines 560 - 564, nimRows builds a table from nimCatalog.models but doesn't guard against missing minGpuMemoryMB, which can produce NaN; update the mapping in nimRows to check m.minGpuMemoryMB != null (or !== undefined) and render a safe fallback (e.g., "unknown" or an empty value) when it's missing instead of dividing and calling toFixed on undefined, mirroring the defensive handling used in get_nim_models.
🧹 Nitpick comments (2)
test/mcp-docs-server.test.cjs (1)
202-214: Version test doesn't verify the server's reported version.The test reads
package.jsondirectly and asserts it equals"0.1.0", but it doesn't check that the server actually returns this version in itsinitializeresponse. ThemcpCallhelper always looks for responseid: 2, but the initialize response hasid: 1.Consider either:
- Extracting and verifying
serverInfo.versionfrom theid: 1response- Renaming the test to clarify it only validates the package.json value
♻️ Proposed fix to verify server-reported version
it("server version matches package.json", () => { - const res = mcpCall("initialize", { - protocolVersion: "2024-11-05", - capabilities: {}, - clientInfo: { name: "test", version: "1.0" }, - }); - // The init response is id 1, but we sent it as id 2 in mcpCall wrapper - // Use a direct check instead + // Get the initialize response (id: 1) from the mcpCall output + const init = JSON.stringify({ + jsonrpc: "2.0", + id: 1, + method: "initialize", + params: { + protocolVersion: "2024-11-05", + capabilities: {}, + clientInfo: { name: "test", version: "1.0" }, + }, + }); + const output = execSync(`node "${SERVER}"`, { + input: `${init}\n`, + encoding: "utf-8", + timeout: 15000, + cwd: ROOT, + }); + const response = JSON.parse(output.trim().split("\n")[0]); const pkg = JSON.parse( fs.readFileSync(path.join(ROOT, "mcp-docs-server", "package.json"), "utf-8") ); - assert.equal(pkg.version, "0.1.0", "package.json version mismatch"); + assert.equal(response.result.serverInfo.version, pkg.version, "server version mismatch"); });🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@test/mcp-docs-server.test.cjs` around lines 202 - 214, The test currently reads package.json (pkg) and asserts its version but never checks the server's initialize response (which comes back with id: 1) because mcpCall wrapper expects id: 2; update the test to capture the raw initialize response (call "initialize" and read the response with id === 1 or adjust mcpCall to accept an expectedId) and assert that response.result.serverInfo.version equals pkg.version (or assert response.serverInfo.version depending on the shape), using the existing mcpCall/initialize call and pkg variable to compare the two versions so the test verifies the server-reported version matches package.json.mcp-docs-server/README.md (1)
48-60: Consider removing leading$from console commands.The static analysis tool flags dollar signs used before commands without showing output. While this is a common convention, removing them improves copy-paste usability.
♻️ Proposed fix
```console -$ cd /path/to/NemoClaw -$ node mcp-docs-server/index.js +cd /path/to/NemoClaw +node mcp-docs-server/index.jsThe server communicates over stdin/stdout using the MCP JSON-RPC protocol.
Install Dependencies
-$ cd mcp-docs-server -$ npm install +cd mcp-docs-server +npm install</details> <details> <summary>🤖 Prompt for AI Agents</summary>Verify each finding against the current code and only fix it if needed.
In
@mcp-docs-server/README.mdaround lines 48 - 60, Remove the leading "$"
prompt characters from the shell command examples in README.md so they are
copy-pastable; update the code blocks containing "cd /path/to/NemoClaw", "node
mcp-docs-server/index.js", "cd mcp-docs-server", and "npm install" to omit the
"$" prefixes while leaving the rest of the text and formatting intact.</details> </blockquote></details> </blockquote></details> <details> <summary>🤖 Prompt for all review comments with AI agents</summary>Verify each finding against the current code and only fix it if needed.
Duplicate comments:
In@mcp-docs-server/index.js:
- Around line 560-564: nimRows builds a table from nimCatalog.models but doesn't
guard against missing minGpuMemoryMB, which can produce NaN; update the mapping
in nimRows to check m.minGpuMemoryMB != null (or !== undefined) and render a
safe fallback (e.g., "unknown" or an empty value) when it's missing instead of
dividing and calling toFixed on undefined, mirroring the defensive handling used
in get_nim_models.
Nitpick comments:
In@mcp-docs-server/README.md:
- Around line 48-60: Remove the leading "$" prompt characters from the shell
command examples in README.md so they are copy-pastable; update the code blocks
containing "cd /path/to/NemoClaw", "node mcp-docs-server/index.js", "cd
mcp-docs-server", and "npm install" to omit the "$" prefixes while leaving the
rest of the text and formatting intact.In
@test/mcp-docs-server.test.cjs:
- Around line 202-214: The test currently reads package.json (pkg) and asserts
its version but never checks the server's initialize response (which comes back
with id: 1) because mcpCall wrapper expects id: 2; update the test to capture
the raw initialize response (call "initialize" and read the response with id ===
1 or adjust mcpCall to accept an expectedId) and assert that
response.result.serverInfo.version equals pkg.version (or assert
response.serverInfo.version depending on the shape), using the existing
mcpCall/initialize call and pkg variable to compare the two versions so the test
verifies the server-reported version matches package.json.</details> --- <details> <summary>ℹ️ Review info</summary> <details> <summary>⚙️ Run configuration</summary> **Configuration used**: Path: .coderabbit.yaml **Review profile**: CHILL **Plan**: Pro **Run ID**: `4426a3c0-87ae-47f7-8a15-942c52fad75c` </details> <details> <summary>📥 Commits</summary> Reviewing files that changed from the base of the PR and between fd496d1c1fcb93204b75869350da14bce7b82e43 and c661b07aed4ace88a893dcc7311cd524c75ed48c. </details> <details> <summary>⛔ Files ignored due to path filters (1)</summary> * `mcp-docs-server/package-lock.json` is excluded by `!**/package-lock.json` </details> <details> <summary>📒 Files selected for processing (6)</summary> * `.github/workflows/pr.yaml` * `.gitignore` * `mcp-docs-server/README.md` * `mcp-docs-server/index.js` * `mcp-docs-server/package.json` * `test/mcp-docs-server.test.cjs` </details> <details> <summary>✅ Files skipped from review due to trivial changes (2)</summary> * .gitignore * mcp-docs-server/package.json </details> </details> <!-- This is an auto-generated comment by CodeRabbit for review status -->
…l inference (NVIDIA#209) * feat(inference): add sandbox-system inference route for platform-level inference Add a separate system-level inference route that the sandbox supervisor can use in-process for platform functions (e.g., embedded agent harness for policy analysis), distinct from the user-facing inference.local endpoint. The system route is accessed via an in-process API on the supervisor, ensuring userland code in the sandbox netns cannot reach it. - Extend proto with route_name fields on Set/Get inference messages - Add ResolvedRoute.name field to the router for route segregation - Server resolves both user and sandbox-system routes in bundles - Sandbox partitions routes into user/system caches on refresh - Expose InferenceContext::system_inference() in-process API - CLI --sandbox flag targets the system route on set/get/update - Integration tests using mock:// routes for the full in-process path Closes NVIDIA#207 * refactor(cli): rename --sandbox flag to --system for inference commands The --sandbox flag could be misread as targeting a user-level sandbox operation. Rename to --system to clearly indicate it configures the platform-level system inference route. * style(cli): collapse short if-else per rustfmt 1.94 --------- Co-authored-by: John Myers <johntmyers@users.noreply.github.com>
ac12644
force-pushed
the
feat/mcp-docs-server
branch
from
5c67671 to
7318bad
Compare
|
@cv Rebased and merged with latest main. Also addressed all CodeRabbit feedback (path traversal fix, empty query guard, null checks, Windows path resolution) and added the MCP server install step to CI. Ready for review. |
|
@ac12644 this is a good idea, but since the interfaces for the NemoClaw CLI are still very unstable, I think this work would be better done a little later. I'm going to close this for now, and once the UX is more stable, we can revisit. I'd like to have a |
4 participants
Summary
mcp-docs-server/) that exposes NemoClaw documentation, blueprint config, network policies, NIM model catalog, and source code as 12 searchable tools for AI assistants.mcp.jsonto.gitignore(contains user-specific paths)Tools
list_docsread_docsearchget_blueprint_configget_baseline_policylist_policy_presetsget_policy_presetget_nim_modelsread_source_filelist_source_filesget_architecture_overviewget_dockerfileTest plan
node --test test/mcp-docs-server.test.js).mcp.jsonconfigSummary by CodeRabbit
New Features
Documentation
Tests
Chores