v0.8.0

• Generation no longer times out in MCP clients. Generating a policy takes
  40-60 seconds, and most MCP clients abandon a tool call after 60 seconds
  while the server finishes anyway, so a retry created a duplicate policy.
  generate_policy, generate_baa and regenerate_policy now start a
  background job, wait for it for up to 45 seconds while sending progress
  notifications, and return the result as before when it is ready. If it is
  still running, they return a job ID instead.
• New check_generation tool: returns the result of a job, its error, or
  that it is still running. Tool output tells the agent to use it rather than
  start the generation again.
• get_policy on a new policy that is still being generated reports the job
  instead of "not found" (the job ID is also the new policy's ID).
• Requires the API's ?async=true generation mode and GET /api/v1/jobs/{id},
  both live on policyforge.co. Against a server without them, the tools fall
  back to waiting for the synchronous response.

v0.7.0

• update_policy accepts edits: a list of exact find/replace pairs applied
  to the current content, so a one-line fix no longer means resubmitting the
  whole document. Each find must match exactly once, and if any edit fails
  to match, none are applied. Cannot be combined with content.
• update_policy warns when an update introduces repeated consecutive lines.
• Corrected the update_policy description: every content or title update
  saves the previous version first, so restore_policy_version can undo it.
• generate_policy and regenerate_policy list the claims to verify that the
  API now returns: specific statements about third parties, certifications,
  data transfer frameworks or deadlines that the inputs did not supply, plus
  any marker left for a missing detail such as governing law.
• generate_policy tells agents to ask for governing_law for terms of
  service, EULAs and disclaimers.
• get_integration_guide notes that the hosted link and the embed both make
  visitors' browsers contact policyforge.co, and adds a third option that
  renders the policy on the site's own page from the API.
• Server-side (no package change needed): stack_manifest is accepted again.
  Since late July 2026 every request carrying it was rejected with "stack_manifest must
  be a string". Generated policies no longer state an AI provider's retention
  period, EU-US Data Privacy Framework reliance, a breach notification
  deadline in hours, certifications or encryption standards unless the inputs
  supplied them.

v0.6.1

The server now starts without POLICYFORGE_API_KEY instead of exiting, so a host can list its tools before the user has configured a key. Previously it called process.exit(1) before ever serving tools/list, which meant MCP clients and directory crawlers saw a server that failed to start rather than one awaiting credentials. Calling any tool without a key still fails, now as a normal 401 carrying the same setup guidance.

Adds a Dockerfile so the server can be built and introspected in a clean container, and a glama.json declaring the maintainer.

Install: npx -y @policyforge/mcp@0.6.1

v0.6.0

Tool schemas regenerated from the API field contract, adding table_of_contents — a Pro option that prefixes a generated document with a bulleted list of anchor links to each section.

Server-side improvements needing no package change: privacy policies, terms of service and EULAs gained substantially more clause coverage, including a CCPA/CPRA categories table with per-category retention, the controller/processor split, GPC signals, sub-processor change notification and an EU representative section. Where a third-party AI provider is detected, policies now state explicitly whether customer content trains models and how long the provider retains it.

Install: npx -y @policyforge/mcp@0.6.0