Pathmode
OfficialPathmode MCP server lets you manage IntentSpecs, evidence, and workspace context to find open product decisions before coding.
Read intents: get current, list, search, fetch by ID with relations, and filter by status.
Write intents: create, update, change status, save to intent.md, and export as Cursor rules or CLAUDE.md.
Manage evidence: query, create, and link evidence (friction, quotes, observations, metrics, requests) to intents.
Analyze intent graph: critical path, cycles, bottlenecks, orphans, status mismatches, stalled intents.
Generate agent prompts for drafting or executing an intent, and verify implementations against the spec.
Access workspace context: strategy, constitution rules, products, and full workspace details.
Log implementation notes and export context for AI agents.
@pathmode/mcp-server
Find the product decisions a request leaves open, before your coding agent builds. A deterministic preflight: six calibrated checks, keyless, no model call.
This repository dogfoods the same workflow: read its intent.md.
This repository is the issue tracker for the package, not its source. The server is developed in Pathmode's private monorepo and published to npm from there. This repo previously held a hand-copied snapshot of the source, which drifted several versions behind the published package. It no longer carries that copy, so there is nothing here to read out of date.
To read the code, unpack the published tarball:
npm pack @pathmode/mcp-server && tar -xzf pathmode-mcp-server-*.tgzInstall
Add to .mcp.json in your project root:
{
"mcpServers": {
"pathmode": {
"command": "npx",
"args": ["@pathmode/mcp-server"]
}
}
}No account and no API key. Specs live in intent.md in your repo, as plain markdown you own. Set
PATHMODE_API_KEY to sync intent and evidence across a team.
Related MCP server: sokrates
See it work
Given one instruction, implement FN-412, in Fieldnote
(a fictional billing app whose ticket asks for a Cancel plan button), an agent with this server
wrote no product code. It found that the only cancel code in the repository was written for
refunds and would take paid days away from every customer who cancels, drafted an intent.md with
four open product choices, and stopped: preflight passed 6/6, still blocked by those choices.
The recorded run
lists every tool call, and npx -y @pathmode/cli preflight on that branch reproduces the verdict.
Links
npm package: the published server, always current
IntentSpec: the open specification for the
intent.mdfiles this server writesThe intent.md profile: the six checks
check_intent_readinessscores againstPathmode: the platform for teams
Listed in the MCP registry
Issues
Bug reports and feature requests belong in Issues.
Include the output of npx @pathmode/mcp-server --version and the client you are running.
License
MIT
Available Tools
20 toolsanalyze_intent_graphAnalyze Intent GraphARead-only
Analyze the intent dependency graph for risks and strategic insights. Returns critical path, cycles, bottlenecks, orphans, status mismatches, and stalled intents.
Inputs: analysis.
create_evidenceCreate EvidenceA
Create a new evidence item (e.g., a discovered bug, user feedback quote, behavioral observation, or feature request). Evidence can later be linked to intents to support prioritization.
Inputs: content (required), type (required), productId (required), source, +5 more.
create_intentCreate IntentA
Create a new intent spec in the workspace. Requires at minimum a title, objective, and productId. Returns the created intent with its ID. Use list_intents first to see existing intents and avoid duplicates.
Inputs: title (required), objective (required), productId (required), outcomes, +5 more.
export_contextExport ContextARead-only
Export workspace context as a formatted file. Use "claude-md" for CLAUDE.md (full workspace context), "cursorrules" for Cursor AI rules, or "intent-md" for a single intent specification file. For cursorrules/intent-md, product context is always derived from the resolved intent. For claude-md, pass productId to select a specific product, otherwise the first active product is used.
Inputs: format (required), intentId, productId.
get_agent_promptGet Agent PromptARead-only
Get a formatted execution prompt for a specific intent. This is the full structured prompt including objective, outcomes, constraints, edge cases, and verification steps.
Inputs: intentId (required), mode.
get_constitutionGet ConstitutionARead-only
Get the workspace constitution rules. These are mandatory constraints that all implementations must respect.
Inputs: no named parameters.
get_current_intentGet Current IntentARead-only
Get the currently active intent, preferring approved intents with real objective/outcome content over empty stubs. Returns the full IntentSpec with objective, outcomes, constraints, and edge cases.
Inputs: status.
get_intentGet IntentARead-only
Get a single intent by ID with full details including objective, outcomes, constraints, edge cases, and relations.
Inputs: intentId (required).
get_intent_relationsGet Intent RelationsBRead-only
Get the dependency graph for a specific intent. Shows what it depends on, enables, or blocks.
Inputs: intentId (required).
get_workspaceGet WorkspaceARead-only
Get workspace details including strategy (vision, non-negotiables, architecture principles), active products, and constitution rules.
Inputs: no named parameters.
intent_exportA
Export an intent spec as .cursorrules or CLAUDE.md section for AI agent consumption.
Inputs: format (required), spec (required), path.
intent_saveB
Save an intent spec to intent.md in the project root. Called after building a spec through conversation.
Inputs: spec (required), path.
link_evidenceLink Evidence to IntentA
Link or unlink evidence items to/from an intent. Linking evidence to intents establishes traceability between user problems and planned solutions.
Inputs: intentId (required), link, unlink.
list_intentsList IntentsARead-only
List all intents in the workspace. Returns an array of IntentSpecs with their status, objectives, and metadata.
Inputs: status.
log_implementation_noteLog Implementation NoteA
Record a technical decision or implementation note for an intent. Use this to document why you chose a specific approach.
Inputs: intentId (required), note (required).
query_evidenceQuery EvidenceARead-only
Search evidence items (friction points, user quotes, observations, metrics, feature requests) by product, type, severity, or text. Returns matching evidence with IDs that can be linked to intents.
Inputs: productId, type, severity, search, +2 more.
search_intentsSearch IntentsARead-only
Search intents by keyword. Searches across user goals, objectives, outcomes, and constraints.
Inputs: query (required), status.
update_intentUpdate IntentAIdempotent
Update an existing intent's content. Provide only the fields you want to change. Does NOT change intent status (use update_intent_status for that).
Inputs: intentId (required), title, objective, outcomes, +5 more.
update_intent_statusUpdate Intent StatusAIdempotent
Update the status of an intent. Use this to mark an intent as shipped after implementation, or verified after testing. For shipped/verified transitions, the response includes a verification checklist of outcomes, constitution rules, and health metrics that should be confirmed.
Inputs: intentId (required), status (required).
verify_implementationVerify ImplementationA
AI-grade your implementation against the intent spec. Checks each outcome, constraint, constitution rule, and edge case. Returns pass/fail per item with reasoning and an overall score. Also logs the result as an implementation note.
Inputs: intentId (required), summary (required), codeChanges.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 20 tools
Most tools have distinct purposes (e.g., create_intent vs list_intents; get_intent_relations vs analyze_intent_graph). However, there is some overlap between export_context, intent_export, and intent_save, and between get_intent_relations and analyze_intent_graph, which could cause confusion for an agent.
The majority of tools follow a verb_noun pattern (e.g., create_intent, list_intents, update_intent). However, a few tools break this pattern (e.g., intent_export, intent_save, get_current_intent), introducing minor inconsistency.
With 20 tools, the set is on the heavy side of the typical range. While each tool serves a specific function, some could potentially be consolidated (e.g., intent_export vs export_context). The count is not excessive but warrants a slightly lower score.
The tool set covers most lifecycle operations for intents and evidence (create, read, update, search, link, export). However, there are notable gaps: no tools to delete intents or evidence, and no tool to manage products beyond workspace retrieval.
Maintenance
Related MCP Connectors
Turn PRDs and product ideas into structured specs so coding agents build your intent, not theirs.
Spec-driven delivery: plans, solutions, ADRs, UX docs, tasks, rules and skills for AI agents.
Persistent, governed institutional memory for Claude Code — specs, decisions, learnings.
Source-checked CLI guides and model-aware planning for Claude Code, Codex, and Grok Build.
Related MCP Servers
- AlicenseAqualityCmaintenancePrevents premature AI coding by transforming vague product ideas into structured specifications, architecture decisions, and acceptance criteria through a series of interrogation and compilation tools.723 npmMIT
- FlicenseNot gradedqualityDmaintenanceEnables users to clarify vague goals through Socratic questioning and generate structured technical specifications, using a curated knowledge base and session state without external LLM APIs.-
- AlicenseBqualityBmaintenanceLocal-first Agent OS that wraps Claude Code, Codex CLI, and other coding agents in a replayable Seed → Ledger → Runtime contract, driven by an interview → seed → execute → evaluate → evolve workflow loop.3433,002 PyPI6,381MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI coding agents to run a specification-first workflow with Socratic interviews, immutable seed specs, staged evaluation gates, and persistent evolutionary loops across multiple runtimes.MIT