Coding agent access

Henry can act as an MCP server for coding agents running in your CI pipeline. A coding agent fixing a production bug can list your Vercel deployments and read runtime and build logs through Henry, scoped to your workspace, read-only, without holding your Vercel credentials.

The token

A workspace admin reveals the token in Settings → Team settings → Coding agent access. Store it as the HENRY_AGENT_MCP_TOKEN secret in each repository that needs it.

The token is read-only and reaches nothing else in your workspace.
Regenerate invalidates the current token immediately. Every repo using it must update its secret.
Format: hmcp_<workspace-id>.<signature>.

Prerequisites in Henry

Connect Vercel in Integrations, granting All Projects when Vercel asks. Optionally connect Supabase at team scope for read-only database tools.

Point your agent at Henry

Add Henry to your agent’s MCP configuration:

{
  "mcpServers": {
    "henry": {
      "type": "http",
      "url": "https://app.usehenry.ai/api/agent-mcp",
      "headers": { "Authorization": "Bearer <HENRY_AGENT_MCP_TOKEN>" }
    }
  }
}

Tools appear to the agent under the henry server:

Tool	Arguments
`vercel_list_projects`	none
`vercel_list_deployments`	`project` (name or `prj_` id), optional `limit`
`vercel_get_runtime_logs`	`project`, optional `deploymentId` (defaults to the newest production deployment)
`vercel_get_build_logs`	`deployment` (`dpl_` id or URL)
`SUPABASE_*`	Read queries and schema. Writes are refused unless a workspace admin has armed a time-boxed grant; destructive SQL is always rejected.

The endpoint is stateless JSON-RPC over a single POST and supports initialize, ping, tools/list, and tools/call. Batch requests are rejected.

Smoke test

curl -s -X POST https://app.usehenry.ai/api/agent-mcp \
  -H "Authorization: Bearer $HENRY_AGENT_MCP_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"vercel_list_projects","arguments":{}}}'

Response	Meaning
`401 unauthorized`	Stale or mistyped token. Reveal it again in Settings.
`503 agent MCP not configured`	Coding agent access is not enabled on this Henry deployment.
”No authorized Vercel connection”	Connect Vercel in Integrations.
”no projects visible”	Reauthorize Vercel and choose All Projects.

Workspace skills for the agent

GET https://app.usehenry.ai/api/agent-mcp/skills with the same bearer token returns the workspace’s shared skills as { enabled, generatedAt, context, files }, so an agent can load your team’s conventions before it starts. Personal skills are never included. When skills are off for the workspace it returns enabled: false.

A dispatch pipeline

Teams that dispatch coding agents from chat typically wire a workflow_dispatch GitHub Action that takes a ticket id, title, and description, works on a branch named after the ticket, opens a pull request, and runs an automated review. Henry can trigger such a workflow through your GitHub integration; the dispatch is a write action and stages an approval card like any other. Nothing merges on its own.

Never enable step debugging or full-output logging on workflows that use this token. Run logs would then contain production logs and live query results.

Email support@usehenry.ai for the reference workflow files.

Overview

Extending Henry

Data and security

Coding agent access

The token

Prerequisites in Henry

Point your agent at Henry

Smoke test

Workspace skills for the agent

A dispatch pipeline

​The token

​Prerequisites in Henry

​Point your agent at Henry

​Smoke test

​Workspace skills for the agent

​A dispatch pipeline

The token

Prerequisites in Henry

Point your agent at Henry

Smoke test

Workspace skills for the agent

A dispatch pipeline