PromptHub
Manage, retrieve, and execute versioned prompts and projects through the PromptHub API.
Verdict
Common use cases
- Run versioned customer-support prompts from chat
- Fetch latest product-description template for batch jobs
- List all classification prompts by model provider
- Execute QA prompt with dynamic variables inline
- Retrieve committed prompt config before deployment
Integration
- Vendor
- PromptHub
- Category
- other
- Auth
- API_KEY
- Tools
- 4
- Composio slug
prompt_hub
Tools
- Get Current Team
Return the current PromptHub team ID for the connected account. Use it when a team-scoped operation must target an explicit team.
- Get Project Head
Retrieve a project's latest committed prompt and model configuration, with optional branch and variable resolution.
- List Projects
List projects in a PromptHub team, optionally filtered by group, model, or provider.
- Run Project
Execute a committed PromptHub project prompt with optional variables and return the generated text, usage, cost, and timing metadata. This invokes the configured LLM, consumes the PromptHub plan's LLM request quota, and creates persistent r
Setup
Setup guide
- 11. In Switchy, open Settings → Integrations and click Connect next to PromptHub. 2. You'll be prompted for a PromptHub API key; generate one at prompthub.us/settings/api-keys with read and execute permissions. 3. Paste the key and click Save. 4. Switchy confirms the connection by fetching your team ID. 5. Open any Project in Switchy and type '@PromptHub list projects' to see your available prompts. 6. To test execution, pick a project name from the list and run '@PromptHub run project <name>' — the MCP calls PromptHub's LLM, returns the generated text, and logs token usage. 7. If you work across multiple PromptHub teams, note that the key is scoped to one team; reconnect with a different key to switch contexts.
What teammates see: by default, memories from PromptHub are scoped to the Project (PROJECT visibility) - you can mark any memory PRIVATE or share it ORG-wide.
Works well with
Top models
Compatibility data appears once enough Projects have used this MCP together with a given model.
How Switchy teams use it
Starter prompts
List All Projects
@PromptHub list all projects in my team and show me the name, model, and provider for each.
Run Support Triage
@PromptHub run the 'support-triage' project with the variable 'ticket_body' set to the last message in this thread.
Fetch Latest Prompt
@PromptHub get the latest committed version of the 'product-description' project and show me the full prompt text.
Filter by Provider
@PromptHub list all projects that use OpenAI as the provider and tell me which models they're configured for.
Run with Variables
@PromptHub run 'email-draft' with variables 'recipient_name' as 'Alex' and 'topic' as 'Q4 roadmap review'.
Example outputs
Illustrative - representative of the model's voice and quality, not literal recordings.
@PromptHub list all projects in our team, filtered by the GPT-4 model, so I can see which prompts are using it.
Found 7 projects in your PromptHub team configured with GPT-4: 'Customer Onboarding Email', 'Support Ticket Classifier', 'Product Description Generator', 'Meeting Summary Template', 'Code Review Assistant', 'Sales Outreach Draft', and 'Bug Report Triage'. Each project includes its group assignment, provider (OpenAI), and last commit timestamp. This list reflects your team's current prompt library as of the most recent sync.
This example shows PromptHub's role as a centralized prompt registry. Useful when auditing which workflows depend on a specific model or provider. Requires a valid API key with team-read permissions; the list reflects committed projects only, not drafts.
@PromptHub run the 'Support Ticket Classifier' project with the variable ticket_text set to the last message in this thread.
Executed 'Support Ticket Classifier' (GPT-4, main branch). Classification result: 'Billing Issue – High Priority'. The run consumed 340 tokens (280 prompt, 60 completion), cost $0.0204, and completed in 1.8 seconds. This invocation used your PromptHub plan's LLM quota and returned the model's raw output along with usage metadata for cost tracking.
This write-action example demonstrates PromptHub's ability to execute versioned prompts with dynamic variables. Each run consumes your PromptHub LLM quota and incurs cost. Ensure the project exists and variables match the template's schema, or the run will fail with a validation error.
@PromptHub get the latest committed version of our 'Meeting Summary Template' project and show me the prompt text and model config.
Retrieved 'Meeting Summary Template' from main branch (commit abc123f). Prompt: 'Summarize the following meeting transcript in 3 bullet points, focusing on action items and decisions: {{transcript}}'. Model: GPT-4-turbo, temperature 0.3, max tokens 500. This is the production version currently deployed; any changes in draft branches are not reflected here until committed.
This read-heavy example showcases PromptHub's version-control layer for prompts. Useful when you need to audit or replicate a prompt's exact configuration. The response shows only committed state; draft edits require a branch parameter. No LLM invocation occurs, so no quota is consumed.
Use-case deep-dives
When PromptHub works for engineering teams shipping LLM features in internal tools
A 4-engineer team is adding LLM summarization to an internal dashboard—meeting notes, Slack threads, ticket descriptions. They want to iterate on the summarization prompt without redeploying the dashboard, and they want usage/cost tracking baked in. PromptHub's Run Project tool executes the prompt and returns cost metadata; Get Project Head pulls the latest version so the team can edit prompts in PromptHub's UI and see changes instantly. The limit: if you need to chain multiple LLM calls (summarize, then extract action items, then route to Slack), the 4-tool cap means you're managing orchestration outside PromptHub. For single-step LLM features at early-stage scale, it's a practical way to keep prompt logic separate from application code and get cost visibility without building it yourself.
Frequently asked
What does the PromptHub MCP do in Switchy?
It lets your AI agents pull versioned prompts from PromptHub and run them directly inside Switchy conversations. Instead of hardcoding prompts in your workflow, you store them in PromptHub, commit changes like code, then call them by project name. The MCP handles variable substitution and returns the LLM output plus cost and timing metadata.
Do I need a PromptHub API key to connect it?
Yes. The MCP uses API key authentication, so you'll need to generate one in your PromptHub account settings and paste it into Switchy during setup. The key is scoped to your PromptHub team, so anyone on that team can see the projects the MCP lists and runs.
Can the PromptHub MCP create or edit prompts?
No. It's read-and-execute only. You can list projects, fetch the latest committed version, and run prompts with variables, but you can't commit new versions or change configurations through the MCP. Use PromptHub's web UI or Git workflow to manage prompt content.
Why use this instead of just copying prompts into Switchy?
Version control and team reuse. When you update a prompt in PromptHub, every Switchy agent calling that project gets the new version automatically. You also get branch support, rollback, and a single source of truth for prompt engineering across your team's workflows.
Does running a PromptHub project count against my Switchy usage?
The MCP call itself is free, but running a project invokes the LLM configured in PromptHub and consumes your PromptHub plan's LLM quota, not Switchy's. The MCP returns cost and token usage so you can track spend per execution.