Your agent's prompts should not live in source code.
Agents need prompts that can be updated without redeploying, tested without guessing, and composed from reusable parts. That's a prompt management problem.
How Promptmark fits
Playbooks with @prompt expansion
Reference any prompt in your library from a playbook step using @prompt:library/slug. When the playbook runs, the prompt is fetched, its template variables are filled, and the rendered content becomes the step instruction. Update the prompt in your library and every playbook that references it picks up the change.
MCP client connections
Connect your Promptmark instance to external MCP servers — giving your playbooks access to tools from other services. Your agent doesn't just use prompts; it can call external APIs, query databases, or trigger actions through MCP-connected tools.
Trigger URLs and delivery
Every playbook can be exposed as an HTTP endpoint. Your agent framework calls the trigger URL with input variables. The playbook runs its steps, branches on conditions, and delivers results to a webhook, email, or GitHub repo. No Promptmark SDK required.
BYOK across providers
Your agent might use different models for different tasks. Promptmark connects to 300+ models through your own API keys. Switch models per step, per playbook, or per test.
66 MCP tools for agent-native access
Your agent framework manages its own prompt library through MCP — create prompts, render templates, list versions, run safety scans. Authenticated via OAuth 2.0 with device flow.
Version control for safe prompt updates
When you update an agent's prompt, the previous version is preserved automatically. Restore the previous one and the next trigger URL call uses it.
Template variables for dynamic agent inputs
Define typed parameters on agent prompts. Schema validation catches malformed inputs before they reach the model. One prompt template serves every invocation.
Example workflow
Build the prompt library
Create prompts for each agent capability: classify-intent, generate-response, summarize-thread. Add template variables for dynamic inputs.
Wire them into a playbook
Create a playbook that chains the prompts. Use @prompt:library/classify-intent to reference each prompt. Add branching based on classification results.
Test the full chain
Run the playbook with sample inputs. Check branching, output flow, and delivery. Try different models for each step.
Deploy the trigger URL
Your agent framework calls the playbook's trigger URL. Update any prompt in the library and the next execution uses the new version — no redeploy.
Manage agent prompts, not agent code
Version, test, and deploy your agent's prompts from one library. Update without redeploying.
Build your agent's prompt library — free