Uh oh!
There was an error while loading. Please reload this page.
Jminnetian/sanitize tool names - #1423
Conversation
WalkthroughMCP tool names are sanitized for provider compatibility, while raw names are propagated through streamed metadata and restored in chat tool, approval, and search displays. ChangesMCP tool name sanitization and display mapping
Estimated code review effort: 3 (Moderate) | ~20 minutes Sequence Diagram(s)sequenceDiagram
participant MCPToolSets
participant createAgentStream
participant ChatThread
participant ToolComponents
MCPToolSets->>createAgentStream: expose sanitized tool names and raw-name mappings
createAgentStream->>ChatThread: emit data-mcp-tool model/raw pairs
ChatThread->>ToolComponents: provide McpToolNameContext
ToolComponents->>ToolComponents: render resolved display names
Suggested reviewers: 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (1)
packages/web/src/ee/features/chat/mcp/mcpToolSets.test.ts (1)
117-236: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick winAdd a test for punctuation-collision sanitization.
The new "keeps valid hyphenated tool names distinct from underscored names" test (Lines 220-235) only exercises a non-colliding case (
-is preserved). Consider adding a test with two raw tool names that both sanitize to the same string (e.g.foo.barandfoo_bar) to lock in the intended collision-handling behavior once it's implemented — see the related comment inmcpToolSets.ts.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@packages/web/src/ee/features/chat/mcp/mcpToolSets.test.ts` around lines 117 - 236, Add a collision-focused test in sanitizeMcpToolNameForModel/getMcpTools coverage: the current hyphen-vs-underscore case only proves distinct valid names stay distinct. Create a test with two raw MCP tool names that sanitize to the same model-facing key (for example punctuation vs underscore), and assert the intended collision-handling behavior in getMcpTools, including the resulting tools map and any disambiguation strategy referenced by mcpToolSets.ts.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@CHANGELOG.md`:
- Line 35: The new CHANGELOG.md entry under [Unreleased] is missing the required
pull request link. Update the entry so the single-sentence description is
followed by the matching
[#<id>](https://github.com/sourcebot-dev/sourcebot/pull/<id>) link, consistent
with the surrounding CHANGELOG entries and the project’s CHANGELOG formatting
rules.
In `@packages/web/src/ee/features/chat/mcp/mcpToolSets.ts`:
- Around line 118-127: Sanitized MCP tool names can collide, causing later
entries to overwrite earlier ones in both allTools and toolDisplayNames. Update
the MCP tool registration flow in mcpToolSets.ts where qualifiedName is passed
through sanitizeMcpToolNameForModel so collisions are detected and handled
instead of silently overwriting; if two raw names sanitize to the same key,
disambiguate the generated key (or otherwise preserve uniqueness) before storing
into allTools and toolDisplayNames. Make sure the fix is applied consistently in
the tool-building logic around the qualifiedName handling so each MCP tool keeps
its own execute function and display name.
---
Nitpick comments:
In `@packages/web/src/ee/features/chat/mcp/mcpToolSets.test.ts`:
- Around line 117-236: Add a collision-focused test in
sanitizeMcpToolNameForModel/getMcpTools coverage: the current
hyphen-vs-underscore case only proves distinct valid names stay distinct. Create
a test with two raw MCP tool names that sanitize to the same model-facing key
(for example punctuation vs underscore), and assert the intended
collision-handling behavior in getMcpTools, including the resulting tools map
and any disambiguation strategy referenced by mcpToolSets.ts.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: f74ea55e-abc3-42fc-b22a-fb331728c6f4
📒 Files selected for processing (13)
CHANGELOG.mdpackages/web/src/ee/features/chat/agent.test.tspackages/web/src/ee/features/chat/agent.tspackages/web/src/ee/features/chat/components/chatThread/chatThread.tsxpackages/web/src/ee/features/chat/components/chatThread/detailsCard.tsxpackages/web/src/ee/features/chat/components/chatThread/toolApprovalBanner.tsxpackages/web/src/ee/features/chat/components/chatThread/tools/mcpToolComponent.test.tsxpackages/web/src/ee/features/chat/components/chatThread/tools/mcpToolComponent.tsxpackages/web/src/ee/features/chat/components/chatThread/tools/toolSearchToolComponent.tsxpackages/web/src/ee/features/chat/mcp/mcpToolSets.test.tspackages/web/src/ee/features/chat/mcp/mcpToolSets.tspackages/web/src/ee/features/chat/mcpServerIconContext.tsxpackages/web/src/features/chat/types.ts
| - Upgraded `nodemailer` to `^9.0.1`. [#1356](https://github.com/sourcebot-dev/sourcebot/pull/1356) | ||
| - Upgraded `@opentelemetry/core` to `^2.8.0`. [#1413](https://github.com/sourcebot-dev/sourcebot/pull/1413) | ||
| - [EE] Fixed connector setup dialogs to add scrolling when connector setup content goes out of view. | ||
| - [EE] Fixed Ask connector MCP tools with provider-invalid names failing to run by sanitizing model-facing tool names while preserving raw names in the UI. |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win
Missing PR link in CHANGELOG entry.
Unlike the surrounding entries, this new line has no trailing [#<id>](url) link. As per coding guidelines, "Update CHANGELOG.md with an entry under [Unreleased] linking to the new PR" and "CHANGELOG.md entries must follow the format: single sentence description followed by a link in the format [#<id>](https://github.com/sourcebot-dev/sourcebot/pull/<id>)."
📝 Proposed fix
-- [EE] Fixed Ask connector MCP tools with provider-invalid names failing to run by sanitizing model-facing tool names while preserving raw names in the UI.+- [EE] Fixed Ask connector MCP tools with provider-invalid names failing to run by sanitizing model-facing tool names while preserving raw names in the UI. [`#1423`](https://github.com/sourcebot-dev/sourcebot/pull/1423)📝 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.
| -[EE] Fixed Ask connector MCP tools with provider-invalid names failing to run by sanitizing model-facing tool names while preserving raw names in the UI. | |
| -[EE] Fixed Ask connector MCP tools with provider-invalid names failing to run by sanitizing model-facing tool names while preserving raw names in the UI.[`#1423`](https://github.com/sourcebot-dev/sourcebot/pull/1423) |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@CHANGELOG.md` at line 35, The new CHANGELOG.md entry under [Unreleased] is
missing the required pull request link. Update the entry so the single-sentence
description is followed by the matching
[#<id>](https://github.com/sourcebot-dev/sourcebot/pull/<id>) link, consistent
with the surrounding CHANGELOG entries and the project’s CHANGELOG formatting
rules.
Source: Coding guidelines
| /** | ||
| * Provider APIs such as OpenAI Responses enforce tight charset and length | ||
| * limits for tool names. MCP tool names are server-controlled, so normalize the | ||
| * fully qualified name before exposing it to the model. | ||
| */ | ||
| export function sanitizeMcpToolNameForModel(name: string): string { | ||
| const sanitized = name.replace(/[^A-Za-z0-9_-]/g, '_') || '_'; | ||
| return sanitized.slice(0, MODEL_TOOL_NAME_MAX_LENGTH); | ||
| } | ||
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Sanitized tool names are not collision-free.
sanitizeMcpToolNameForModel replaces every disallowed character with _, so two distinct raw tool names can map to the same sanitized key — e.g. catalog.query and catalog_query both become catalog_query. Truncation to 64 chars can also collide names sharing a long common prefix. Since qualifiedName is used to key both allTools (Line 330) and toolDisplayNames (Line 277), a collision silently overwrites the earlier tool's execute function and permission wiring with the later one's — the earlier tool becomes unreachable and its display name is lost, with no error or log.
Given MCP tool names are server-controlled and can contain arbitrary punctuation, this isn't a purely theoretical edge case.
🛠️ Proposed fix: detect and disambiguate collisions at the call site
const originalExecute = tool.execute;
const rawQualifiedName = `${prefix}__${toolName}`;
- const qualifiedName = sanitizeMcpToolNameForModel(rawQualifiedName);+ let qualifiedName = sanitizeMcpToolNameForModel(rawQualifiedName);+ if (qualifiedName in allTools) {+ let suffix = 1;+ let candidate = qualifiedName;+ do {+ const suffixStr = `_${suffix++}`;+ candidate = `${qualifiedName.slice(0, MODEL_TOOL_NAME_MAX_LENGTH - suffixStr.length)}${suffixStr}`;+ } while (candidate in allTools);+ qualifiedName = candidate;+ }
const timeoutMs = env.SOURCEBOT_MCP_TOOL_CALL_TIMEOUT_MS;
toolDisplayNames[qualifiedName] = toolName;Also applies to: 274-277
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@packages/web/src/ee/features/chat/mcp/mcpToolSets.ts` around lines 118 - 127,
Sanitized MCP tool names can collide, causing later entries to overwrite earlier
ones in both allTools and toolDisplayNames. Update the MCP tool registration
flow in mcpToolSets.ts where qualifiedName is passed through
sanitizeMcpToolNameForModel so collisions are detected and handled instead of
silently overwriting; if two raw names sanitize to the same key, disambiguate
the generated key (or otherwise preserve uniqueness) before storing into
allTools and toolDisplayNames. Make sure the fix is applied consistently in the
tool-building logic around the qualifiedName handling so each MCP tool keeps its
own execute function and display name.
| @@ -8,3 +8,10 @@ export type McpServerIconMap = Record<string, string>; | |||
| export const McpServerIconContext = createContext<McpServerIconMap>({}); | |||
| export const useMcpServerIconMap = () => useContext(McpServerIconContext); | |||
| // Maps provider-safe model tool names back to raw MCP tool names for display. | |||
| export type McpToolNameMap = Record<string, string>; | |||
There was a problem hiding this comment.
nit: consider moving this into a separate file
a3c6855 to
0ccbb4bCompareThere was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 0ccbb4b. Configure here.
Uh oh!
There was an error while loading. Please reload this page.
MCP server tool names with certain characters (e.g " . ") would break the tool names open AI was expecting with their models. We now sanitize the names to avoid running into this issue.
Summary by CodeRabbit
Note
Low Risk
Targeted EE Ask/MCP naming and display path with tests; execution still uses raw MCP names, so risk is limited to display/metadata and model-facing identifiers.
Overview
Ask connector MCP tools whose names include characters providers reject (e.g. dots in
catalog.query-catalog-entities) were failing at the model layer. This change normalizes fully qualified tool keys exposed to the LLM viasanitizeMcpToolNameForModel(invalid chars →_, max 64 chars) while MCP execution and persistence still use the server’s raw tool name.The agent now streams
data-mcp-toolparts mappingmodelToolName→rawToolName. The chat thread hydrates that intoMcpToolNameContext(alongside server favicons in the renamedmcpDisplayMetadataContext). Tool cards, approval banners, and connector tool search usegetMcpToolDisplayPartsso users see names likebackstage: catalog.query-catalog-entitiesinstead of the sanitized model identifier; older threads without metadata keep the previous parse-based fallback.Tests cover sanitization,
getMcpToolsbehavior, agent streaming, and UI rendering.Reviewed by Cursor Bugbot for commit 2edd846. Bugbot is set up for automated code reviews on this repo. Configure here.