back to docs

Skill Server

Expose your ChatBotKit skillsets as a text-first HTTP API that agents read and invoke directly - a simpler alternative to MCP that needs no client.

The Skill Server integration turns any ChatBotKit skillset into a self-describing, text-first HTTP API. Instead of speaking the Model Context Protocol, an agent simply reads a plain-text manual that lists the available abilities and how to call them, then invokes those abilities directly with a single HTTP request. It is the sibling of the MCP Server integration - same skillset, simpler transport.

What You Can Do

  • Publish Skillsets as a Plain API: Take any skillset and expose its abilities over a simple HTTP endpoint
  • No Client Required: Any agent framework, script, or curl can drive it - there is no MCP client, no JSON-RPC, and no handshake
  • Self-Describing: A generated manual documents every ability and its inputs, so agents discover capabilities on their own
  • Text-First: Responses are plain text by default (token-cheap and truncation-friendly), with JSON available on demand
  • Secure Access: Each integration gets its own static access token so you control who can call your abilities

How It Works

When you create a Skill Server integration and attach a skillset, ChatBotKit exposes two authenticated endpoints: a manual endpoint that returns a Markdown description of the abilities, and an invoke endpoint that executes a single ability by name. Both are authenticated with the integration's static access token as a bearer token.

An agent reads the manual once to learn what is available, then calls invoke with the ability name and its input. ChatBotKit executes the ability and returns the result as plain text (or JSON when ?format=json is set).

Getting Started

  1. Create the Integration: Give it a name and click Create
  2. Select a Skillset: Choose an existing skillset that contains the abilities you want to expose
  3. Save: Click Save to generate your endpoints and access token
  4. Copy the Configuration: Expand the Installation Instructions section for ready-to-run curl examples and the endpoint URLs
  5. Point Your Agent at It: Have the agent read the manual, then invoke abilities as needed

Best Practices

Keep Skillsets Focused: Each integration exposes one skillset. Purpose-specific skillsets keep the manual short and make it easier for an agent to pick the right ability.

Protect Your Access Token: The token unlocks every ability in the skillset, including any that touch secrets, bots, spaces, or files. Store it in a secure vault and inject it via environment variables.

Name Abilities Clearly: The manual is generated from your ability names and descriptions. Clear names and descriptions help the agent choose correctly.

Practical Use Cases

Lightweight Agent Tooling: Give a custom agent framework access to your abilities without implementing an MCP client.

Scripts and Automation: Call abilities from shell scripts, cron jobs, or CI pipelines with a single curl.

Knowledge Retrieval: Expose a search skillset so an agent can look up company knowledge over plain HTTP.