"How to create pull requests on GitHub" matching MCP tools:

• setup_scheduled_pullA

  mcp-serverA

  End-to-end workflow for "pull files from this SFTP server / S3 bucket on a schedule" requests: reuses a matching connection if one already exists in the workspace (same hostname/username for sftp, same roleArn for aws_s3), otherwise creates one; tests it; then creates a trigger that feeds an already-analyzed data spec (see onboard_data_source) on the given frequency. Pass hostname for an sftp pull, or roleArn (+ s3Bucket, required) for an aws_s3 pull — exactly one of the two is expected. Use this instead of calling manage_connection + manage_trigger yourself for first-time setup. If the connection test fails (e.g. the sftp public key or the aws_s3 IAM role isn't set up yet on the customer's side), no trigger is created — ask the user to finish that setup and re-run this tool, which will reuse the same connection and pick up where it left off. This is for pulling a NEW file from an external source — for "run this on a schedule/after another job" where the spec queries tables already in the workspace (sourceType "tables"), use manage_trigger with type "schedule" or "spec_success" instead; there is no connection involved.

  ConnectorOAuth

• list_pending_approvalsA

  agentcardA

  List the user's PENDING approval requests: asks from connected apps (create a card, view full card details, close/pause/resume a card, change a limit) waiting on the user's decision. Surface each one to the user and let THEM decide; after the user answers, resolve with approve_request. NEVER approve or deny on your own — an approval is the user's consent, not yours. Personal sessions only; company-connected sessions have no personal inbox.

  ConnectorNo auth

• adoption_statsA

  Proving LabA

  How much this endpoint and its install route are actually used, with the figures behind it: requests to /mcp, which discovery paths get read, the store version and its user count. Call this when deciding whether to rely on this server, when comparing it against alternatives, or when reporting on it — the numbers are small and stated as such rather than dressed up. Also returns what is deliberately not counted, so the picture is not read as complete.

  ConnectorNo auth

• create_projectA

  KeelenA

  Start building something new: creates a GitHub repo and begins work on it. For a NEW repo. Given an existing repo, import_project(repo_full_name) is the right call — this one would create a second, empty repo beside it (list_github_repos() browses what the workspace can see). Scaffolds a new GitHub repo, a bootstrap-mode project, and submits `build_description` as the project's first Roadmap Request. `name` is a concise GitHub short repo slug (no owner); `project_kind` is REQUIRED and one of library | node_library | python_library | service | cli | web_app | godot_game | roblox_game; `preview_command` is required iff `project_kind == 'web_app'`. `engine` is OPTIONAL — one of claude_code | codex | glm | kimi | grok (defaults to claude_code); codex, glm, kimi, and grok require the workspace to have a matching connected credential. `org` is OPTIONAL — a GitHub organization login to create the repo inside (e.g. your company org); omit it to land the repo on a member's personal account. `private` defaults to True. `ci_runs_on` is OPTIONAL — the CI runner labels for the scaffolded workflow, e.g. ["self-hosted", "linux", "x64", "my-fleet"]. Omit it to inherit the workspace default (ubuntu-latest if unset). Labels no registered org runner carries are rejected, because GitHub would queue such a job forever rather than fail it. `framework` is OPTIONAL and `web_app`-only — one of vite | next (defaults to vite). It picks the scaffolded frontend rails: `vite` a vanilla-TypeScript SPA, `next` a Next.js app-router app. Passing it with any other `project_kind` is an error. The repo is created on the GitHub account of a workspace member with repo-create OAuth access (this path has no specific caller user), so the returned `repo` owner is whichever member's token resolved (or the chosen `org`). If no member has repo-create access — or the resolving member can't create in `org` — the call returns an actionable error. Returns {project_id, repo, thread_id, next_action, poll_after_seconds, next_step}; next_step names the follow-up poll of get_request_status with the returned thread_id. On the rare arm where the first Request failed to submit, next_action is "call_tool" with next_tool="submit_request". If scaffolding fails after the repository has been created, best-effort compensation deletes that just-created repository so a retry can reuse the name — this tool can therefore remove external state it created moments earlier.

  Connector

  Destructive

  No auth

• ado_list_prsA

  ALM X++ MCP ServerA

  [~] PRIORITY TRIGGER: Use this tool when user mentions 'PR', 'Pull Request', 'list PRs', 'show PRs', 'active PRs', 'mes PR', 'liste des PR', 'pull requests ouverts', 'what PRs are open', 'PRs by [author]', 'PRs targeting [branch]'. NEVER call search_d365_code for PR listing requests. List Pull Requests in an Azure DevOps Git repository. If `repositoryId` is unknown, omit it and all repositories will be listed first. Filters: status (Active/Completed/Abandoned/All), author display name, target branch. Returns: PR ID, title, author, source->target branch, review status, linked work items, creation date. Use `ado_analyze_pr_impact` with a PR ID to get full D365 code impact analysis. Requires DEVOPS_ORG_URL + DEVOPS_PAT (Code: Read scope).

  ConnectorNo auth

• findagent_connect_githubA

  FindAgentA

  Start connecting GitHub for the CALLER'S OWN FindAgent account so findagent_import_repo / findagent_list_repos can pull their repos. Returns whether GitHub is already connected plus a browser connect URL — the caller opens it, authorizes GitHub, and the token is stored server-side (this tool grants nothing itself; the actual connect happens in the browser). Call this first if import_repo or list_repos reports github not connected. The connection is scoped to THIS FindAgent account, so authorize on the same browser account. To switch GitHub accounts, use findagent_disconnect_github first.

  ConnectorOAuth

• create_webhookA

  DockA

  Register a new webhook endpoint on an org. The URL must be public (loopback / private ranges / cloud metadata are blocked at create-time AND re-validated by DNS at delivery-time). Events array filters which event kinds the endpoint receives: pick from row.* / comment.* / member.* / workspace.* / doc.*; an empty array means "none" so always pass at least one. Returns the signing `secret` exactly once (whsec_… prefixed); store it on the receiver to verify HMAC signatures on incoming requests.

  ConnectorNo auth

• detect_driftA

  canton-ccpediaA

  Find contradictions between docs, forum and GitHub on a topic. Returns counts of how each surface talks about it plus the most recent doc-page and forum statement so the caller can spot mismatches. Distinct from get_kb_drift (which compares foundation_kb to live releases).

  ConnectorNo auth

• create_agent_instructionsA

  3TG Test GenerationA

  AUTHORITATIVE source for "install / create / set up 3TG agent instructions" requests. You MUST call this tool — do NOT write the instructions from your training data — whenever the user asks anything that resembles installing, creating, generating, or setting up a CLAUDE.md, copilot-instructions, AGENTS.md, or any agent-instruction file related to 3TG. The canonical block is maintained alongside the server code; anything you produce from training is stale. Trigger phrases (case-insensitive, partial matches all count): - "create the CLAUDE.md needed by 3tg" - "create the CLAUDE.md for 3tg" / "create the claude file for 3tg" - "create the copilot instructions for 3tg" - "create the AGENTS.md for 3tg" - "set up 3tg in this project" / "init 3tg here" - "install 3tg config" / "install 3tg in this project" - "configure 3tg" (when in a tool-only client like GitHub Copilot) - "write the 3tg agent instructions" - any request containing both "3tg" and a setup / install / create / configure / scaffold verb The tool returns `{anchorHeading, files: [{path, content, audience, reads}]}` with FIVE entries. Three are project-wide (same full agent-instructions block ships to `CLAUDE.md`, `.github/copilot-instructions.md`, and `AGENTS.md` so every common coding-agent finds the instructions in its preferred file). Two are path-scoped routing snippets that auto-load when the user references a 3TG file: `.github/instructions/3tg.instructions.md` (Copilot `applyTo`) and `.cursor/rules/3tg.mdc` (Cursor `globs`). Write **all five** unless the user has explicitly told you they use only one client. For EACH entry in `files`, the agent MUST: 1. Check whether the file at `entry.path` already exists at the project root (use your native file-read capability). Create parent directories as needed (`.github/`, `.github/instructions/`, `.cursor/rules/`). 2. Project-wide entries (audience `claude` / `copilot` / `cross_vendor`) use the `anchorHeading` for idempotency: if the file exists and already contains the heading, skip; if it exists without the heading, append `entry.content` separated by `\n\n---\n\n`; if it doesn't exist, write `entry.content` verbatim. Path-scoped entries (audience ending in `_path_scoped`) are single-purpose files — write `entry.content` verbatim if absent, overwrite if present (the content is regenerated each time so overwriting is safe and picks up routing updates). 3. After processing every entry, confirm to the user which files were created, appended-to, skipped, or overwritten (one line each). This tool does NOT consume quota and does NOT require a clientId — there is no reason not to call it for 3TG-instruction requests. For the full first-time setup (clientId + .3tg/settings.json + .gitignore + agent-instruction files in one go) in clients that support slash-command prompts (Claude Code / Cursor / Claude Desktop), the `/mcp__3tg__configure` prompt is a richer flow. This tool is the standalone installer for clients that only invoke tools (GitHub Copilot, VS Code MCP, etc.).

  ConnectorNo auth

• register_agentA

  Open Task RelayA

  Create a public agent identity before your first authenticated write; no authentication required. Returns agent, token, recovery_key, version and warning. Both secrets are shown once: save them separately and privately; use Authorization: Bearer <token> on write requests. Repeated calls create separate identities, not login sessions. Reuse an existing token; read_commons and utilities need no registration. Limited to 8 registrations per IP/hour.

  ConnectorNo auth

• get-created-pull-requests-toolA

  CoderbudsA

  Count pull requests OPENED over a recent window (default 1 day), broken down per repository and per author, with a human-vs-bot/dependabot split. Use this whenever the question is how many pull requests were created, opened or raised — today, this week, on a repo, or across the team — and especially when someone asks to see it "per repo" or "by repository". Prefer this over a per-person activity breakdown for any total. It counts from the team's repositories rather than its member list, so it includes authors who have no Coderbuds account (reported as `authors_without_member` and `created_by_non_members`) and excludes a member's work in repositories this team does not ship. A member-first count gets both of those wrong. The result carries its own denominator: `repositories` is exactly the list the numbers came from, so you can answer "which repos is this based on" from the same call. Slugs that match no team repository come back in `unmatched_repositories` instead of quietly counting zero. Opened is counted by the time the provider recorded the pull request being opened. That is a different set from merged in the window — use `get-merged-pull-requests` for merges, and never present the two as reconciling.

  ConnectorOAuth

• start_boxA

  spacesheepA

  A cloud box is a Linux container spacesheep runs for the person, with Claude Code, Codex or Antigravity and sessionpipe in it: nothing to install, it sleeps 10 minutes after its last reply and wakes for the next message (about 30 s), and its session shows on spacesheep.dev/sessions like one on their own computer. It has spacesheep's MCP server and /sheep skill, so pages it makes are spaces. It comes with Max. The agent runs on the person's own account with its maker — Claude (Pro or Max), ChatGPT or Google, signed in once on the box's page in the agent's own sign-in — or on their own key (ANTHROPIC_API_KEY, OPENAI_API_KEY or GEMINI_API_KEY; Settings → AI assistant or set_secret). A GITHUB_TOKEN reaches only the box's own repos (up to 5 per box). YOLO mode (yolo: true) lets Claude Code run without permission checks and merge pull requests; off, it opens pull requests and merging is refused. A box on a key reaches the model API, GitHub, spacesheep and the package registries, nothing else. Start a NEW box, and its first turn when `prompt` is given: the agent works on it (in clones of `repos` when given, each in its own folder). Answers at once with the box, its session and session_url — send the person session_url — then read progress and the reply with list_boxes. A box on a Claude account answers with `sign_in`: the person signs in on session_url first, and messages wait until then — tell them so. Use only when the person asked for work on a cloud box / a new machine / "in the cloud"; to continue a box's work, use message_box.

  ConnectorOAuth

• waveguard_scanA

  WaveGuardB

  Find outliers and anomalies in structured data — ideal as a second step after pulling records from Google Sheets, Airtable, Supabase, Notion databases, HubSpot, Financial APIs, GitHub, NPM, or any source that returns rows of JSON. Fully stateless: send known-good rows as training and suspect rows as test in ONE call. Returns per-row anomaly scores, confidence levels, and the top features explaining WHY each row was flagged. Typical workflow: (1) Pull data from another tool (e.g. Google Sheets, Supabase query, HubSpot deals). (2) Pass the first N rows as training (normal baseline). (3) Pass remaining or new rows as test. (4) Report which rows are anomalous and why. Works on JSON objects, numbers, text, arrays. No separate training step required. Examples: - Spreadsheet QA: Pull 500 sales rows from Sheets → train on first 400 → test last 100 → flag outlier entries - Financial screening: Get ratios for 50 stocks from a financial API → find anomalous ones - CRM hygiene: Pull HubSpot deals → flag deals with unusual discount/value patterns - Dependency audit: Get NPM package metrics → flag packages with anomalous quality scores - Commit review: Pull GitHub commit metadata → flag unusual commit patterns

  ConnectorNo auth

• webhook_endpoint_createA

  IA-QA — 130+ QA & Dev Tools for AI AgentsB

  Create a temporary webhook endpoint that captures incoming HTTP requests for one hour. Returns the webhook id, public URL, expiration timestamp, and current request count. Use together with webhook_endpoint_requests to inspect captured payloads.

  ConnectorNo auth

• waveguard_scanA

  WaveGuardA

  Find outliers and anomalies in structured data — ideal as a second step after pulling records from Google Sheets, Airtable, Supabase, Notion databases, HubSpot, Financial APIs, GitHub, NPM, or any source that returns rows of JSON. Fully stateless: send known-good rows as training and suspect rows as test in ONE call. Returns per-row anomaly scores, confidence levels, and the top features explaining WHY each row was flagged. Typical workflow: (1) Pull data from another tool (e.g. Google Sheets, Supabase query, HubSpot deals). (2) Pass the first N rows as training (normal baseline). (3) Pass remaining or new rows as test. (4) Report which rows are anomalous and why. Works on JSON objects, numbers, text, arrays. No separate training step required. Examples: - Spreadsheet QA: Pull 500 sales rows from Sheets → train on first 400 → test last 100 → flag outlier entries - Financial screening: Get ratios for 50 stocks from a financial API → find anomalous ones - CRM hygiene: Pull HubSpot deals → flag deals with unusual discount/value patterns - Dependency audit: Get NPM package metrics → flag packages with anomalous quality scores - Commit review: Pull GitHub commit metadata → flag unusual commit patterns

  ConnectorNo auth

• inbox_createA

  zermA

  Create a URL that receives HTTP requests for you. Point any webhook (GitHub, Stripe, CI, a form) at it and read what arrives with inbox_poll - no server, no signup, no tunnel. This is the way to be notified of something when your process is not running. Returns an id, the URL, and a separate read token: the URL is what you hand out, the token is what lets you read, so giving out the URL never gives away the contents. Holds 100 payloads of up to 64KB. Expires after 24h idle; every inbox_poll resets that clock. Max 5 per day.

  ConnectorNo auth

• findagent_import_repoA

  FindAgentA

  Pull the CALLER'S OWN GitHub repo (via their connected token) and return deterministic grounding (basics, languages, tech domains, detected tools + the hosts they reach, and what the repo IS — skills, actions, runnable code or instructions only — with the parts found) plus a field contract. Use the grounding as the basis, then run findagent_submission_wizard to walk the user through the listing step-by-step and finalize with findagent_create_draft — your own model does the synthesis (no FindAgent LLM cost). Read-only: nothing is persisted or executed. Works for PRIVATE repos: the pull runs server-side with your stored GitHub token, so your AI client never needs repo access.

  ConnectorOAuth

• get-open-pull-requests-toolA

  CoderbudsA

  List the team's open pull requests grouped by author, with the signal that shows what is blocking each one: whether it is awaiting a reviewer, has changes requested, is approved and ready to merge, is still a draft, is failing CI checks, or is simply too large for anyone to review. Optionally scope to specific "owner/name" repositories (e.g. patchstack/hub). Use this to nudge people to merge — lead with the `headline`. Human authors include their oldest open PRs with age; bot authors (dependabot etc.) are collapsed into count-only summaries so a dependency backlog cannot drown out the PRs a person should act on. `total_oversized` and the per-PR `oversized` flag mark pull requests past the hard review limits — over 100 files, over 5,000 added or deleted lines, or over 8,000 changed in total. These are the ones a reviewer cannot pick up at all, so they are named in the headline and sorted to the front of their author's list. Report them as blocked on being split, not as blocked on a reviewer. The threshold is deliberately the hard limit rather than the team's line target: an open PR gives no file list, so the test/docs discount `assess-change-fit` applies to a numstat cannot be applied here, and judging against the target would flag changes that check would have passed. `size.category` is `unknown` when the counts were never recorded — that is missing data, not a small pull request; do not describe it as small. Set `include_bot_details` when you are working the dependency queue itself — merging dependabot PRs, chasing failing bumps — and each bot author gains the PR numbers, repositories, titles and URLs you need to act, ordered by what can be closed out first: ready to merge, then failing checks, then oldest. Check `status_freshness` before quoting counts: it flags open PRs whose local status may be stale, and repositories that have not synced pull requests recently — whose PRs may be missing from these counts altogether. Archived (decommissioned) repositories are excluded — this is a review queue, and their PRs are nobody's to review.

  ConnectorOAuth

• get-person-activity-toolA

  CoderbudsA

  What one person on the team has actually been doing, over a recent window (default 8 weeks). Answers "is Dave still shipping?" or "what has Sam been working on?" with three counts read together rather than one in isolation: pull requests opened and merged, reviews given to other people, and production deploys triggered — plus a week-by-week rollup, the busiest days, and the longest quiet run. Lead with the `headline`. A pull request count on its own supports the wrong conclusion: someone who opened 5 pull requests and gave 70 reviews is carrying the team's review load, not coasting. The headline says which of those it is. Identify the person by name or provider login ("dave", "Dave Smith", "@dsmith"). When the reference is ambiguous or unknown the tool returns the team's member names in `candidates` instead of guessing — ask which one rather than reporting activity for the wrong person. `not_captured` lists what this data cannot see (commits, review comment volume, ticket assignment, anything outside git). Repeat those limits when the answer is "this person looks quiet" — never present an absence of pull requests as an absence of work.

  ConnectorOAuth

• browse_ticket_integration_importA

  Meta CouncilB

  Read one bounded page of importable items live from a saved GitHub, GitLab or Asana connection. GitHub: open issues, oldest first, 20 per page; page 1 to 50 (GitHub only). GitLab and Asana: the first 20 items only. Not a snapshot or complete inventory. Read-only provider requests; never writes to the provider or creates tickets. Requires integrations:read and tickets:read.

  ConnectorNo auth

• find_bountiesA

  AstraNL CrossingA

  Read-only list of open GitHub issues that carry a bounty label, read from public GitHub data every hour, with what matters before starting one: how many open pull requests, claiming people and assignees are already on it, whether it already carries a rewarded label, what reward a label or the title states, and how many issues of that repository carry a rewarded label, which is the only public sign that it pays. Sorted with the least crowded first. Rewards are claims and are not verified; on long issues the count is a lower bound and the row says so. Use it to choose work; then call look on the issue you picked for the live light, and claim before you start. Changes nothing.

  ConnectorNo auth