How do I use Highway MCP?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Highway MCP run_goal 'summarize files in /tmp'" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

Highway MCP

by rodmena-limited

Overview Schema Source Related Servers Discussions Admin

Python

Hybrid

Highway MCP

The MCP front door for Highway Agents. It lets any MCP host (Claude Desktop/Code, opencode, Cursor, Claude mobile) drive durable, always-on agents that run server-side on the Highway Workflow Engine — the user runs no infrastructure, just an MCP client and a Highway API key.

It exposes three tools, each a thin wrapper over the Highway REST API:

Tool	What it does	REST call
`run_goal(goal)`	Start a durable agent run; returns `workflow_run_id`. The agent keeps running even if the client disconnects.	`POST /api/v1/workflows`
`get_status(workflow_run_id)`	Status, progress, result, and `pending_approvals`.	`GET /api/v1/workflows/<id>` + `GET /api/v1/approvals`
`approve(approval_key, approved, comment)`	Resolve a human-in-the-loop approval to resume the run (works across devices/sessions).	`POST /api/v1/approvals/<key>/approve\|reject`

The agent loop itself is the engine tool tools.agent.run_goal driven by the agent_run_goal workflow (one durable model turn per iteration, HITL approval before outbound actions). See the engine repo (enterprise/tools/agent.py, api/dsl_templates/agent_run_goal.py) and issue #749.

Prerequisites

The Highway stack running and reachable (default http://localhost:7822).
A Highway API key (hw_k1_...) with permissions submit_workflows, view_workflows, view_approvals, approve_workflows. Mint one via POST /api/v1/admin/api-keys (see engine docs).

Related MCP server: agentic-orchestrator MCP server

Run

pip install -e .

# stdio (for a local host like opencode/Claude Desktop)
HIGHWAY_BASE_URL=http://localhost:7822 HIGHWAY_API_KEY=hw_k1_xxx highway-mcp

# remote / streamable-http (reachable by laptop + mobile clients)
TRANSPORT=http PORT=8848 HIGHWAY_BASE_URL=https://highway.rodmena.app \
  HIGHWAY_API_KEY=hw_k1_xxx highway-mcp

Connect from opencode (dogfood)

Add to the host's MCP config (stdio example):

{
  "mcpServers": {
    "highway-agents": {
      "command": "highway-mcp",
      "env": {
        "HIGHWAY_BASE_URL": "http://localhost:7822",
        "HIGHWAY_API_KEY": "hw_k1_xxx"
      }
    }
  }
}

Then: run_goal("summarise what files are in /tmp") → the agent asks to run a shell command → get_status(<id>) shows a pending approval → approve(<key>) → the run resumes and completes; get_status returns the result.

Configuration

Env var	Default	Purpose
`HIGHWAY_BASE_URL`	`http://localhost:7822`	Engine base URL
`HIGHWAY_API_KEY`	(required)	`hw_k1_...` key, forwarded as Bearer
`TRANSPORT`	`stdio`	`stdio` or `http`/`streamable-http`
`PORT`	`8848`	HTTP port when `TRANSPORT=http`

Status / follow-ups (Phase 0 spike)

Auth is single-key from env. Production needs per-connection auth so each customer's own key (from the MCP session/OAuth) is forwarded — not one shared server key.
Tool surface is a placeholder (shell_run → tools.shell.run, approval required). Validate the argument contract live and replace with the real connector tools (Gmail etc.) in Phase 1.
Long-running calls already use the start→poll pattern (run_goal returns a handle; get_status polls). Progress-notification/resource streaming is a later nicety.
Engine-side durability caveat for side-effecting tools: issue #750.

Available Tools

4 tools

approveA

Approve (or reject) a pending agent action to resume the durable run.

Set approved=False to reject. This works across sessions and devices: you can approve from one client an action a run started from another.

ParametersJSON Schema

Name	Required	Description	Default
`approval_key`	Yes
`approved`	No
`comment`	No

Output Schema

ParametersJSON Schema

Name	Required	Description
No output parameters

TDQS

A4/5.0

Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations are provided, so the description carries the burden. It reveals that setting approved=False rejects the action and that the tool works across sessions/devices. However, it does not mention potential side effects, error conditions (e.g., invalid approval_key), or idempotency, leaving gaps in transparency.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is concise: two sentences. The first sentence states the main purpose, and the second adds cross-session capability. No redundant or unnecessary information.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the sibling tools and presence of an output schema (not shown), the description covers core functionality and cross-session behavior. It does not specify how to obtain approval_key or what 'pending' means, but these can be inferred from context. Reasonably complete for a simple tool.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The input schema has 3 parameters with 0% description coverage. The description provides meaning for approved (set false to reject) but does not explain approval_key format or the comment parameter. It partially compensates but lacks full detail.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's purpose: 'Approve (or reject) a pending agent action to resume the durable run.' It uses specific verbs (approve/reject) and identifies the resource (pending agent action). This distinguishes it from sibling tools like get_status, run_goal, and schedule_goal.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explains when to use the tool (to approve/reject a pending action) and notes it works across sessions/devices. It does not explicitly mention when not to use it or compare to alternatives, but the context is clear enough for an AI agent to select it over siblings.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

get_statusA

Get an agent run's status, progress, result, and any pending approvals.

pending_approvals lists actions the agent has paused on; pass an approval_key to the approve tool to let it continue.

ParametersJSON Schema

Name	Required	Description	Default
`workflow_run_id`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
No output parameters

TDQS

A4.1/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Without annotations, the description carries full burden. It discloses that the tool returns status, progress, result, and pending_approvals, and explains that pending_approvals are actions the agent paused on. No destructive behavior is implied, and it is clear that this is a read operation.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is extremely concise: two sentences that first state the purpose and then provide key behavioral detail about pending_approvals. No filler words, good front-loading.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

With one required parameter and an output schema present, the description covers the return values (status, progress, result, pending_approvals) and the relationship to the 'approve' tool. It could mention error cases but is adequate for its simplicity.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters2/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 0%, but the description does not elaborate on the 'workflow_run_id' parameter beyond its name. The parameter is self-explanatory as an ID, but the description adds no additional context or formatting guidance.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states it retrieves status, progress, result, and pending approvals for an agent run. It distinguishes from siblings: 'approve' is for approving actions, 'run_goal' and 'schedule_goal' are for execution, while this tool is for retrieval.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description implies when to use: to check run status and pending approvals. It mentions that pending approvals can be acted on with the 'approve' tool, providing context for usage. However, it does not explicitly state when not to use this tool.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

run_goalA

Start a durable Highway agent that pursues goal to completion.

Returns immediately with a workflow_run_id. The agent runs server-side and survives this client disconnecting. Use get_status to follow it and to see any actions awaiting your approval.

ParametersJSON Schema

Name	Required	Description	Default
`goal`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
No output parameters

TDQS

A4.1/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Discloses key behaviors: immediate return, durability, server-side execution, survival of disconnection. No annotations exist, so description carries full burden; lacks error handling or auth details but is sufficient.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Two sentences, each earning its place: first defines purpose, second explains behavior and next steps. No redundancy.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given low complexity (one parameter, output schema exists), the description covers the essential workflow. Could mention the output schema explicitly, but it's not missing critical information.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters2/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

With 0% schema description coverage, the description fails to elaborate on the goal parameter beyond its name. No format, examples, or constraints provided, leaving the agent to infer valid inputs.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool starts a durable Highway agent that pursues a goal to completion. It distinguishes from sibling tools like schedule_goal and get_status by focusing on immediate execution.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Provides clear guidance: returns a workflow_run_id, runs server-side, suggests using get_status for follow-up. Does not explicitly exclude alternatives or mention prerequisites.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

schedule_goalA

Schedule an agent to run goal repeatedly on a cron schedule (UNATTENDED).

cron_expression is standard 5-field cron in UTC, e.g. '0 9 * * ' = daily 09:00, '/30 * * * *' = every 30 minutes. The scheduled agent runs WITHOUT per-run human approval (creating the schedule is the authorization), so only schedule goals you trust. Uses Highway's durable cron (survives restarts; no history bloat). Returns the schedule details.

ParametersJSON Schema

Name	Required	Description	Default
`goal`	Yes
`cron_expression`	Yes

Output Schema

ParametersJSON Schema

Name	Required	Description
No output parameters

TDQS

A4.8/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description fully covers behavioral traits: it explains that the agent runs unattended, that scheduling serves as authorization, and that it uses durable cron (survives restarts, no history bloat). This goes beyond basic functionality.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is concise and front-loaded: the first sentence states the core purpose. Subsequent sentences add critical details without fluff. Every sentence serves a purpose.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's moderate complexity (2 parameters, cron scheduling), the description covers usage, parameters, behavioral nuances, and return value. It is complete for an agent to select and invoke the tool correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters5/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The description explains the 'cron_expression' parameter in detail with format (5-field cron UTC), examples, and meaning. The 'goal' parameter is clear from context. Since schema coverage is 0%, the description compensates effectively, adding significant meaning.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's action: 'Schedule an agent to run `goal` repeatedly on a cron schedule (UNATTENDED).' It specifies the verb 'schedule' and resource 'goal', and the 'UNATTENDED' qualifier distinguishes it from sibling tools like 'run_goal' (which likely runs once with oversight).

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides context that scheduling is for unattended runs and warns that only trustable goals should be scheduled because there is no per-run approval. This implies when to use but does not explicitly compare to alternatives like 'run_goal' or 'get_status', leaving some ambiguity about when to schedule vs. run manually.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

4 tool updatesv0.1.0Jun 21, 2026
- First observedapprove
- First observedget_status
- First observedrun_goal
- First observedschedule_goal

Deploy Server

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4.2/5.0

Scored 2026-06-21 02:50 across 4 tools

Disambiguation5/5

Each tool has a clearly distinct purpose: 'approve' handles pending actions, 'get_status' tracks run progress, 'run_goal' starts a new run, and 'schedule_goal' sets up recurring runs. No overlap or ambiguity.

Naming Consistency4/5

Three tools follow a clear verb_noun pattern (get_status, run_goal, schedule_goal), but 'approve' omits the noun. The pattern is mostly consistent and readable, with a minor deviation.

Tool Count5/5

Four tools is well-scoped for the server's purpose of managing durable agent runs with approvals. Each tool is essential and the set is concise without being sparse.

Completeness3/5

The tools cover starting, scheduling, checking status, and approving runs, but missing cancellation or deletion capabilities (e.g., cancel_run, delete_schedule) are notable gaps that agents may need.

Maintenance

ActivityInactive

ResponsivenessNo issues

Resources

GitHub Repository

Need Help?

Related Servers

Related MCP Connectors

mcp
com.skopik
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
Agent Orchestration Knowledge & Memory Search
2
Agent Router C
io.github.Tim-Projekt
Discover and call AI agents via MCP. Supports A2A agents and platform agents with async tasks.
Agent Orchestration Autonomous Agents
252
Wever Labs Agentic Rails A
com.weverlabs
Agentic rails for complex workflows with receipts, fees, and MCP tool access.
Agent Orchestration Autonomous Agents Payments & Billing
52
aXOAuthDeprecated
io.github.ax-platform
Agent-native collaboration network: orchestrate a team of long-running agents from any MCP client.
Knowledge & Memory Autonomous Agents Agent Orchestration
36

Related MCP Servers

dat-ai-mcp
Browser Automation AI & Machine Learning Speech Processing
willtholke
A
license
A
quality
D
maintenance
Provides browser automation, audio transcription, and LLM chat as MCP tools for any agent.
Updated 3 months ago (2026-07-04 05:04 UTC)
7
MIT
agentic-orchestrator MCP
Agent Orchestration Autonomous Agents
Saul4432
A
license
A
quality
C
maintenance
Enables any MCP client to drive a multi-agent orchestration engine with planning, specialist tools, critic revision, and human-in-the-loop approval for sensitive actions.
Updated 2 months ago (2026-08-01 22:20 UTC)
3
MIT
AIOrc MCP Server
Agent Orchestration Autonomous Agents
Dich01
A
license
Not graded
quality
A
maintenance
Enables MCP-compatible LLM clients to execute server-verified agent workflows, with enforced transitions, invocation caps, and signed audit trails.
Updated 2 months ago (2026-08-06 13:40 UTC)
MIT
ChatGPT Gateway MCP Nyan
Shell Access Command Line Virtualization
wada-qualia
A
license
Not graded
quality
B
maintenance
Enables agent clients to safely connect to tools and execution resources through MCP with authorization, approvals, audit, chat-context isolation, SSH/Docker access, and long-running command session tracking.
Updated 9 days ago (2026-09-14 20:27 UTC)
MIT