Developer tools let the agent bridge customer chat with the engineering stack — file issues from bug reports, look up PR / build status, fetch code to ground answers, trigger deploys, and read runtime logs when debugging.
GitHub
Tool id: github. Credential type: github. Auth: Personal Access Token.
26 actions across six groups: repo info, issues (CRUD + comments), PRs (CRUD + merge + comments), repository contents (get_file, list_directory, create_or_update_file, search_code), workflow runs (list / get / dispatch / cancel), releases / tags, and commits.
Setup — Personal Access Token
- Open github.com/settings/tokens.
- Decide between Fine-grained tokens (recommended — per-repo + per-permission scoping) and Classic tokens (broader).
- For a fine-grained token: select the repo(s), then grant the permissions you need — typical agent scopes are Contents: Read (or Read+Write), Issues: Read+Write, Pull requests: Read+Write, Workflows: Read+Write.
- Generate the token, copy it immediately (starts with
github_pat_for fine-grained orghp_for classic). - Save as a GitHub credential. Set optional default_owner and default_repo so the LLM doesn't have to supply them every call.
- Read-only profile + Contents: Read + Issues: Read = an answer-bot that grounds in code.
- Read+write profile + Issues: Write + Pull requests: Write = the agent can triage and respond.
- Full access profile + Workflows: Write = the agent can dispatch + cancel CI runs.
Use cases
- Triage to engineering — agent calls
create_issuewith the chat transcript as the body andlabels: ["bug", "triage"]. - Status lookups — “is PR #1234 merged?” →
get_pr. “Did last night's deploy succeed?” →list_workflow_runsfiltered by workflow + branch. - Code Q&A — agent calls
get_fileorsearch_codebefore answering API-shape questions instead of guessing. - Trigger a deploy —
dispatch_workflowondeploy.ymlwithinputs.environment = 'staging'.
Vercel
Tool id: vercel. Credential type: vercel. Auth: Access Token.
15 actions across projects, deployments (list / get / create / cancel / promote), build & runtime logs, env vars (list / create / update / delete), and domains.
Setup
- Open vercel.com/account/tokens.
- Click Create. Name it (e.g. chatzuri-agent), pick a scope: full account or a specific team. For team access, also note the team id (in the team URL or under Team Settings → General, ID starts with
team_). - Pick an expiration. Copy the token immediately — shown once.
- Save as a Vercel credential. If team-scoped, also paste the team_id. Optional default_project is the project name or id the agent uses when one isn't supplied.
Use cases
- “Where's the preview?” →
list_deploymentsfiltered by branch and target. The dominant ask. - Build health —
list_deployments target=production limit=1+get_build_logsif it failed. - Runtime debugging —
get_runtime_logswithsince/untiltimestamps. - Promote to prod —
promote_deploymenton a verified preview.
list_env_vars returns metadata only — never the decrypted values. That's deliberate: the LLM should never see your env secrets. If a customer asks “what's the value of DATABASE_URL?”, the right answer is to point them at the Vercel UI, not to print it.Connecting the two
The killer combo is a single agent with both tools wired up: a support-engineering bridge that opens a GitHub issue with the transcript and a link to the latest Vercel preview URL all in one reply.
