Loading ToolBench

Reading the current corpus…

contrib-radar MCP server · C · ToolBench

contrib-radar

AI-powered radar for real OSS contribution opportunities: scores good-first-issue candidates by actual current availability (assignee, competing/abandoned PRs, staleness, maintainer sentiment), not just a label.

2026-07-28+STREAMABLE_HTTP5 tools

67/100

Definition Quality79

Protocol Readiness78

Supportability38

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · evidence: Streamable HTTP

Current-spec patterns detected

Summary

Five tools with clear verb-noun naming (list_, add_, refresh_, check_) and solid descriptions (100-200 chars). All tools have input schemas with Zod validation and parameter descriptions. Error handling uses isError:true pattern. Main gaps: no output schemas documented, no pagination support for list tools, missing parameter constraints (e.g., labelFilter array size limits), and no tool annotations (readOnlyHint/destructiveHint). Composition is clean, each tool has one responsibility. Naming is consistent and action-oriented.

Tools (5)

add_watched_repowriteauthsource verified83/100

Start watching a repo for scored issues - idempotent, returns the existing row if this owner/name is already watched.

check_issue_feasibilityread onlyauthsource verified80/100

Score one issue live, right now, with no caching and no cooldown - the single call that answers "is this issue actually still available, and does a maintainer actually want it worked on?" for a repo that may not even be watched yet.

list_scored_issuesread onlyauthsource verified77/100

List the current scored snapshot for a watched repo, highest-opportunity first by default.

list_watched_reposread onlyauthsource verified78/100

List every repo this API key's owner is watching.

refresh_reporead onlyauthsource verified78/100

Refresh scored issues for a watched repo - a 5-minute cooldown returns the cached snapshot instead of re-hitting GitHub/Gemini.

Issues (5)

HIGHresponse-shaper ↗

No output schemas documented. LLMs cannot plan downstream calls or extract fields without knowing response structure. list_watched_repos, list_scored_issues, and add_watched_repo should document returned fields (id, owner, name, score, etc.).

list_watched_reposadd_watched_reporefresh_repolist_scored_issues

HIGHpaginated-result ↗

list_watched_repos and list_scored_issues lack pagination (limit, offset, total_count). Without pagination, large result sets blow context windows and violate pattern:paginated-result.

list_watched_reposlist_scored_issues

MEDIUMtool-annotations ↗

No tool annotations (readOnlyHint, destructiveHint, idempotentHint). add_watched_repo is idempotent per description but lacks idempotentHint annotation. refresh_repo and check_issue_feasibility are read-only but lack readOnlyHint.

add_watched_reporefresh_repocheck_issue_feasibility

Recommendations

Document output schemas for all tools. Example for list_watched_repos: 'Returns array of {id: string, owner: string, name: string, labelFilter: string[], createdAt: ISO8601, lastRefreshed: ISO8601}'.
Add pagination to list_watched_repos and list_scored_issues: accept limit (1-100, default 20) and offset parameters; return total_count and has_more fields.
Add tool annotations: mark add_watched_repo with idempotentHint:true, refresh_repo and check_issue_feasibility with readOnlyHint:true.
Enhance error handling: catch specific error types (GitHubNotFoundError, RateLimitError, AuthError) and return structured errors with recovery hints. E.g., 'Repo not found. Verify owner/name are correct. Call list_watched_repos to see available repos.'
Add constraints to labelFilter: describe as 'Array of 1-10 GitHub issue labels (max 50 chars each). Defaults to ["good first issue"] if omitted.'

Score history

Overall score trend

First recorded score · v2 rubric

67/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	67	2026-07-28+	v2